Mosaic Analysis with Double Markers reveals IGF1R function in granule cell progenitors during cerebellar development

Abstract

During cerebellar development, granule cell progenitors (GCPs) proliferate exponentially for a fixed period, promoted by paracrine mitogenic factor Sonic Hedgehog (Shh) secreted from Purkinje cells (PCs). Dysregulation of Shh signaling leads to uncontrolled GCP proliferation and medulloblastoma. Serendipitously our previous work discovered insulin-like growth factor 1 (IGF1) as another key driver for medulloblastoma, which led to the current investigation into the role of IGF1 in GCPs during normal development. While the IGF1R conditional knockout model revealed GCP defects in anterior cerebellum, the posterior cerebellum was mostly intact, likely owing to incomplete excision of floxed alleles. To circumvent this hurdle, we enlisted a mouse genetic system called Mosaic Analysis of Double Markers (MADM), which sporadically generates homozygous null cells unequivocally labeled with GFP and their wildtype sibling cells labeled with RFP, enabling phenotypic analysis at single-cell resolution. Using MADM, we found that loss of IGF1R resulted in a 10-fold reduction of GCs in both anterior and posterior cerebellum; and that hindered S phase entry and increased cell cycle exit collectively led to this phenotype. Genetic interaction studies showed that IGF1 signaling prevents GCP cell cycle exit at least partially through suppressing the level of p27kip1, a negative regulator of cell cycle. Finally, we found that IGF1 is produced by PCs in a temporally regulated fashion: it is highly expressed early in development when GCPs proliferate exponentially, then gradually decline as GCPs commit to cell cycle exit. Taken together, our studies reveal IGF1 as a paracrine factor that positively regulates GCP cell cycle in cooperation with Shh, through dampening the level of p27 to prevent precocious cell cycle exit. Our work not only showcases the power of phenotypic analysis by the MADM system but also provides an excellent example of multi-factorial regulation of robust developmental programs

Introduction

Cell number regulation is critical for proper organ development and function. To ensure balanced proportions of different cell types in an organ, paracrine signaling, through which a signaling cell type produces a factor to induce a change in a responding cell type, plays an instrumental role. The cerebellum, a brain region that controls motor coordination and learning, is composed of intricate interactions between excitatory and inhibitory neurons. After integrating all types of inputs, Purkinje cells (PCs) as the sole output from the cerebellar cortex send axons to connect with excitatory neurons in the deep nuclei, which then send outputs toward other brain regions (Sillitoe and Joyner 2007). One of the major input cell types for PCs is granule cells (GCs) (Lackey et al., 2018). The wiring between PCs and GCs is based on the stereotypic organization of three distinct layers from outside to inside: the molecular layer (ML) where GC axons known as parallel fibers form synapses on PC dendrites, the purkinje cell layer (PCL) where PC cell bodies form a monolayer, and the internal granule layer (IGL) where GC cell bodies reside. The connections between PCs and GCs in the ML are particularly remarkable: on one hand, the dendritic tree of a single PC receives inputs from numerous parallel fibers; on the other hand, each parallel fiber can contact a vast number of PCs forming thousands of synapses on their dendritic spines. Conceivably, the GC to PC ratio could be a critical factor to ensure proper wiring and cerebellar function.

Cerebellar development is unique in comparison to other brain regions. While other brain regions develop in an inside-out fashion from neural stem cells (NSCs) residing in the ventricular zone (VZ) of the neural tube (Nadarajah and Parnavelas, 2002), the cerebellum has two germinal zones. While PCs and other cerebellar cell types originate from the VZ and follow an inside-out developmental route (Yuasa et al., 1991); (Leto et al., 2006); (Hoshino et al., 2005), GCs derived from granule cell progenitors (GCPs) in the external granule layer (EGL) and follow an outside-in developmental route (Fig. 1A). In mice, EGL is formed by GCPs derived from a small subset of NSCs that migrate through the rhombic lip (RL) beginning at embryonic day 13.5 (E13.5) that eventually populate the outer surface of the cerebellar primordium (Ben-Arie et al., 1997); (Wang et al., 2005); (Machold and Fishell, 2005). Upon arrival to the EGL, GCPs remain relatively quiescent and only enter a phase of exponential proliferation after birth, producing more than half of the total number of neurons in the brain within two weeks. The postnatal development of GCPs follow a stereotypic process: first GCPs proliferate in the outer EGL (oEGL), then they exit the cell cycle and migrate into the inner EGL (iEGL) and leave behind their axons in the ML as their cell bodies continue to migrate inwardly, and finally they move past the PCL into the IGL and terminally differentiate into GCs (Figs. 1B–D) (Altman and Bayer, 1997); (Cajal et al., 1995); (Roussel and Hatten, 2011).

Figure 1. Schematic illustration of the developmental process of the mouse cerebellum.

(A) The separation of GCP and PC lineages around E14. (B-D) Outside-in development of GCPs from P0 to P21.

Figure 2. Conditional knockout of IGF1R in GCPs led to stunted anterior cerebellar development.

(A) Inactivation of the IGF1R gene results in reduced weight of the cerebellum in comparison to IGF1R WT (+/+) mice (n≥8 for each genotype). (B) Dorsal view of P21 cerebellum from IGF1R WT (+/+), heterozygous (+/Δ), and null (Δ/Δ) mice shows that the decrease in cerebellum size is due to a reduction in the anterior to posterior axis (A-P: dashed line) but not the medial to lateral axis (M-L: solid line). (C-E) Representative H&E images of P21 cerebellum from +/+, +/Δ, and Δ/Δ mice shows stunted development of the anterior cerebellum in Δ/Δ mice in comparison to +/+ and +/Δ mice. (F) The stunted anterior phenotype in Δ/Δ mice is caused by a significant loss of granule cells in the IGL (n=5 per genotype). *VIab is a combination of folia VIa and VIb.

Scale bars: B-E= 2 mm.

Statistics in A, F: one-way ANOVA with Tukey test; error bars ± SEM; ns not significant, *p < .05, **p < .01, ***p < .001.

After the developmental process of PCs and GCs became known, Wetts and Herrup discovered the numerical correlation between these two cell types, known as cell number scaling in the field of developmental biology. Using the lurcher gene mutation that causes PC death (Wetts and Herrup, 1982), they produced chimeric mice with a varied number of healthy WT PCs and found that the number of GCs was positively correlated with the number of PCs (Wetts and Herrup, 1983). While their original interpretation of PC-GC cell number scaling was that the connectivity to PCs ensures GC survival, it could also be alternatively interpreted that PCs provide mitogenic factors that promote GCP proliferation. About 16 years later, it was discovered that PCs secret the mitogenic factor Sonic Hedgehog (Shh) that stimulates GCP proliferation with multiple lines of evidence (Dahmane and Ruiz, 1999); (Wallace, 1999); (Wechsler-Reya and Scott, 1999). Shh expression was detected in the PCL while the Shh responsive gene Gli was expressed by GCPs in the EGL, suggesting that Shh from PCs activates Shh signaling in GCPs (Wechsler-Reya and Scott, 1999). Importantly, Lewis et al. showed that the PC-specific knockout of Shh resulted in the development of a stunted cerebellum with massively reduced GC numbers (Lewis et al., 2004). Taken together, these studies firmly established the critical role of PC-secreted Shh in cell number scaling for GCs.

It should be noted that, in addition to Shh-stimulated GCP proliferation, the timely cell cycle exit of GCPs by the weaning age (postnatal day 21 (P21) in mice) is an important regulatory arm to ensure cell number scaling between PCs and GCs. Due to the large number of GCs (~50% of the total number of neurons in the brain), even just one more or one less cell cycle would lead to a 25% change of total neuron numbers that could lead to brain malformation and disrupted PC-GC wiring. Negative factors for cell cycle progression, such as BMP2, bFGF, and WNT3, have been found to directly antagonize Shh signaling to stop GCP proliferation and promote differentiation (Rios, 2004);(Anne et al., 2013);(Fogarty et al., 2007), although detailed mechanisms still await further investigations. It is particularly puzzling how GCPs exit cell cycle while Shh expression remains high in PCs throughout life.

Due to the importance of Shh for GCP proliferation, uncontrolled Shh signaling leads to medulloblastoma originated from over-proliferated GCPs (Goodrich et al., 1997), (Yang et al., 2008), (Schüller et al., 2008) (Pietsch et al., 1997). Serendipitously, while studying tumor-promoting factors in a mouse model of the Shh-subtype of medulloblastoma, we identified insulin-like growth factor 1 (IGF1) as a critical niche factor for tumor progression (Yao et al., 2020). While this finding indicated IGF1 signaling as another driver of tumor GCP proliferation, we questioned whether IGF1 signaling could also promote the proliferation of normal GCPs, since IGF1 is known to be an important factor for proliferation (Aberg et al., 2003; D’Ercole et al., 1996; DiCicco-Bloom and Black, 1988), differentiation (Hsieh et al., 2004; McCurdy et al., 2005) and survival (Croci et al., 2011; Hodge et al., 2007; Wilkins et al., 2001) of numerous cell types in the central nervous system. Its action is mediated through the IGF1 receptor (IGF1R) and downstream PI3K/Akt and MAPK pathways (Laviola et al., 2007). To investigate the role of IGF1 signaling in the development of normal GCPs in vivo, we first determined whether IGF1 signaling was important for GCP development by inactivating the IGF1R using a conditional knockout model. To overcome the technical problem of inefficient Cre-mediated excision of floxed alleles, next we enlisted a mouse genetic system called Mosaic Analysis of Double Markers (MADM), which sporadically generates homozygous null cells unequivocally labeled with GFP and their wildtype sibling cells labeled with RFP, enabling phenotypic analysis at single-cell resolution. Using MADM, we found that IGF1 is a critical regulator of GC number, and that the loss of IGF1R in GCPs hindered S phase entry and increased cell cycle exit which collectively led to the phenotype. Furthermore, we showed that IGF1 exerts its role in promoting cell cycle progression at least partly by suppressing the level of p27kip1, a CDK inhibitor. Finally, we pinpointed the source cell type for IGF1 is PC, and its temporal expression pattern matched well with GCP development.

Materials and Methods

Mouse Lines

All animal procedures were based on animal care guidelines approved by the Institutional Animal Care and Use Committee at the University of Virginia to ensure the humane treatment in all procedures.

The following mouse lines were used to generate MADM experimental and control mice: MADM-ML pair TG7ML (JAX# 021457, STOCK lis5^{tm2.1(ACTB-EGFP,-tdTomato)Luo}/J), GT7ML (JAX# 021458, STOCK lis5^{tm1(ACTB-tdTomato,-EGFP)Luo}/J), Math1-Cre (JAX# 011104, B6.Cg-Tg(Atoh1-cre)1Bfri/J), IGF1R^flox (JAX# 012251, B6;129-Igf1r^tm2Arge/J), and p27^flox (JAX# 027328, 129S4/SvJae-Cdkn1b^tm2Mlf/J). IGF1R conditional knockout mice were generated by crossing IGF1R^flox with Math1-Cre.

5-Bromo-2’deoxyuridine (BrdU) labeling

BrdU (Sigma, cat. #19–160) was administered by i.p. injection (50mg/kg body weight) for short term labeling (4 hours). Mice were sacrificed after treatment and the brain tissue was collected for analysis.

Tissue preparation and Immunostaining

Adult and perinatal mice were transcardially perfused with cold 4% paraformaldehyde (PFA), then brain tissue was removed, fixed overnight in 4% PFA at 4°C, cryoprotected in 30% sucrose, embedded in optimal cutting temperature (OCT), and stored at −80°C. Brain tissues were cryosectioned sagittally at 20-to-25 μm thickness on glass slides. Slides with brain sections were washed with PBS, then incubated in permeabilizing/blocking buffer (0.3% Triton-X 100 in PBS [PBT] plus 5% normal donkey serum) for 20 min at RT. Primary antibody incubation was performed at 4°C overnight in the same permeabilizing/blocking buffer. Secondary antibody incubation was performed for 2–4 hours at RT in permeabilizing/blocking buffer. To visualize nuclei, slides were incubated in DAPI solution (1ug/mL in PBS) for 5 min before coverslips were mounted with antifade mounting medium. For BrdU immunostaining, slides were treated with freshly prepared 2N HCL (in PBS) at 37°C for 20 min. After acid treatment, slides were washed with PBS, then immunostaining procedures described above were followed. GFP was detected by anti-GFP primary antibody (chicken, 1:500, Cat. no. GFP-1020; Aves Labs). RFP was detected by anti-DsRed primary antibody (rabbit, 1:100, Cat. no. 632496; Takara). Primary antibodies against the following proteins were also used: BrdU (rat, 1:250, Cat. no. OBT0030; Accurate Chemical), cleaved caspase-3 (Asp-175, rabbit, 1:300, Cat. no. 9661S; Cell Signaling Technology), p27kip1 (mouse, 1:100, Cat. no. 610242; BD Transduction Laboratories). Secondary antibodies were conjugated to Alexa Fluor family dyes, purchased from Invitrogen. Fluorescent images were acquired on Zeiss LSM 710 confocal microscope. Confocal images were processed with ImageJ/Fiji and Adobe Photoshop.

Primary cilium staining and quantification

For immunohistochemistry, antigen unmasking was performed by incubating the slides in antigen-retrieval buffer (10 mM Tris Base, 1 mM EDTA, 0.05% Tween-20, pH 9.0) for 5 min at 95 °C. Samples were pre-extracted with 0.3% Triton X-100 in PBS [PBT] for 10 min at room temperature, then incubated in blocking buffer (3.0% BSA in PBT) for 1 hr. Specimens were incubated with primary antibodies overnight at 4 °C at the following dilutions: Arl13b (rabbit, 1:500, Cat. no.17711–1-AP; Proteintech), anti-GFP (chicken, 1:500, Cat. no. GFP-1020; Aves Labs), anti-DsRed primary antibody (rabbit, 1:100, Cat. no. 632496; Takara). Samples were washed in PBT then incubated with Alexa Fluor conjugated secondary antibodies from Invitrogen for 1 hr at room temperature. Nuclei were stained with DAPI, and specimens mounted using Mowiol containing n-propyl gallate (Sigma-Aldrich). Images were captured using a Nikon Eclipse Ti-E inverted confocal microscope. A series of digital optical sections (z-stacks) with 0.200 μm step was captured using an Andor Neo-Zyla sCMOS camera, and three-dimensional image reconstructions were produced using Nikon Elements software. Single cells (either GFP+ or RFP+) were examined from z-stack images to determine the presence or absence of cilia.

In situ hybridization

In situ hybridizations were performed on perinatal cerebellum tissue. RNA probes for IGF1 were prepared from cDNA amplified from postnatal day 8 cerebellum. The amplified fragments were cloned into the pGEMT easy vector (Promega) and sequenced to confirm identity. Digoxigenin-labeled sense and antisense riboprobes were obtained by in vitro transcription using T7 and SP6 transcriptase and digoxigenin-UTP (Roche, Cat. no.11175025910). Perinatal pups were transcardially perfused with cold 4% paraformaldehyde (PFA) prepared in diethyl pyrocarbonate treated PBS, then brain tissue was fixed overnight in 4% PFA at 4°C, cryoprotected in 30% sucrose, embedded in OCT and cryosectioned at 16-μm thickness on glass slides. Hybridization was performed with digoxigenin labeled sense or antisense riboprobes (1 ug/ml) overnight at 62°C in hybridization buffer (50% formamide, 10% Dextran sulfate, 1X salt solution, 1X Denhardt’s solution, 1mg/mL tRNA). The sections were washed with 50% formamide, 1X SSC, 0.1% Tween 20 for 1.5 hr at 62°C then with 1X MABT (final conc. 100 mM of maleic acid, 150 mM of NaCl, deionized water, at pH 7.5 plus 0.1% Tween 20) at room temperature. After blocking with 10% sheep serum, 10% blocking reagent, and 1X MABT, slides were washed with AP buffer (100 mM of Tris, pH 9, 100 mM of NaCl, 50 mM of MgCl2, 0.1% Tween 20) then incubated in a 5-bromo-4-chloro-3-indolyl phosphate (Roche), nitroblue tetrazolium (Roche), and levamisole solution (Vector) to reveal hybridization.

Granule cell progenitor dissociation

GCPs were purified from P8 IGF1R^−/− mice (IGF1R-flox/flox; Math1-Cre) by Percoll gradient subfractionation as described previously (Lee et al., 2009). The cerebellum was dissected in HBSS, the meninges were removed, and the left and right hemispheres were cut away leaving the vermis. Under a dissection microscope the anterior and posterior region of the cerebellum was dissected and collected in separate dishes. Tissue pieces were digested in oxygen-equilibrated enzyme stock solution (final conc. 1XEBSS [Sigma E7510], 0.45% D-glucose, 26mM NaHCO3, 0.5mM EDTA, deionized water) plus 100-Units papain (Worthington, Cat. no. LS003126) and DNase I (final conc. 250ug/mL; Sigma, Cat. no. DN25) at 37°C in 5%CO2 for 45 minutes, and then triturated. After centrifugation the cell suspension was re-suspended in panning buffer (final conc. 0.2% BSA, 1% D-glucose, 2ug/mL DNase I, 1X D-PBS) then passed through a 70-μm cell strainer (Fisherbrand, Cat. no. 22363548). The cell suspension was underlaid with 35% percoll solution followed by 60% percoll solution and centrifuged. Cells at the 35%−60% percoll interface are GCPs and were collected and re-suspended in 1X D-PBS. The purity of GCPs was greater than 95%.

(For 10mL 35% percoll solution: 4mL sterile deionized water, 2.5mL 4X CMF-PBS-EDTA [NaCl 32g/L, KCl 1.2g/L, D-Glucose 8g/L, NaH2PO4.H2O 2g/L, KH2PO4 1g/L, NaHCO3 8ml/L of 2% stock, EDTA 10ml/L of 1M EDTA, pH8.0, deionized water, pH to 7.4 with NaOH, filter sterilize], 3.5mL Percoll [Sigma P4937], Phenol red indicator 5μL, 2N HCl 10μL until solution is pink-orange). (For 10mL 60% percoll solution: 1.5mL sterile deionized water, 2.5mL 4X CMF-PBS-EDTA, 6.0mL Percoll, Phenol red indicator 5μL, 2N HCl 15μL until solution is pink-orange, then add 30μL 0.4% Trypan Blue).

Reverse-transcriptase PCR and qPCR assays

Total RNA was extracted from perinatal cerebellum with TRI Reagent (Sigma, Cat. no. T9424) then treated with DNase I (New England Biolabs Inc, Cat. no. M0303). cDNA was synthesized using iScript Reverse Transcription Supermix (Bio-Rad, Cat. no. 170–8841) from 1 μg of each RNA sample. qRT-PCR was performed with KAPA SYBR FAST ABI Prism Kit (KAPA Biosystems, Cat. no. KK4605) with 10 ng per reaction of DNA, in an Applied Biosystems StepOnePlus Real-Time PCR System for 40 cycles followed by a default Melting Curve program. Target gene expression was normalized to glyceraldehyde-3-phosphate dehydrogenase (Gapdh), or calbindin 1 (Calb1) gene expression. For qPCR, total DNA was extracted from GCPs isolated from P8 cerebellum using DNeasy Blood and Tissue kit (Qiagen, Cat. no. 69504). qPCR was performed with KAPA SYBR FAST ABI Prism Kit (KAPA Biosystems, Cat. no. KK4605) with 25 ng per reaction of DNA, in an Applied Biosystems StepOnePlus Real-Time PCR System for 40 cycles followed by a default Melting Curve program. Each run was performed in duplicate. Target gene expression was normalized to interferon beta 1 (IFNb1) gene expression. Cycle threshold (Ct) values were measured within the geometric amplification phase and averaged for duplicate reactions.

Primers used include:

IGF1 (F; 5’-TGGATGCTCTTCAGTTCGTG- 3’, R; 5’-GTCTTGGGCATGTCAGTGTG -3’)

Gapdh (F; 5’-CAGGTTGTCTCCTGCGACTT-3’, R; 5’-ATGTAGGCCATGAGGTCCAC-3’)

Calb1 (F; 5’- GAGCTATCACCGGAAATGAA-3’, R; 5’- AATTCCTCGCAGGACTTCAG-3’)

IGF1R flox (F; 5’-GAACTAGTGGATCCGTCGAG-3’, R; 5’-AGCTATCGAATTCCTGCAGG-3’)

IFNb1 (F; 5’-CCCTATGGAGATGACGGAGA-3’, R; 5’-CTGTCTGCTGGTGGAGTTC-3’)

Quantification

Granule cell expansion (Fig. 3H; Fig. S3) was quantified from P21 cerebella of MADM-Control and MADM-IGF1R mice. Unbiased sampling was performed by counting all green and red cells in folia II, V, VI, VIII, X from five nonadjacent 25-μm thick sagittal sections (250-μm apart) across the mediolateral span of the cerebellum (six brains total per genotype). Granule cell expansion (Fig. 5C) was quantified from P21 cerebella of MADM-IGF1R, and MADM-IGF1R-p27-Null- mice by counting all red and green cells in the anterior folia (II-VI) of five nonadjacent 25-μm thick sagittal sections (250-μm apart) across mediolateral span of the cerebellum (five brains total per genotype). All quantifications are representative of ≥ 150 GFP+ cells/brain. Cell division rate (Fig. 4D) was determined by counting green and red cells labeled with BrdU, and total green and red cells in the oEGL. About 30–50 green cells per P4 MADM-IGF1R cerebella were counted (eight brains total). BrdU+GFP+ and BrdU+RFP+ cells divided by the total number of GFP+ and RFP+ cells was used to calculate the percentage of dividing cells. Exit fraction (Fig. 4G) was determined by counting green and red cells in the iEGL/ML/IGL divided by the total number of green and red cells (oEGL,iEGL,ML, and IGL). About 50–80 green cells per P4 MADM-IGF1R cerebella were counted (eight brains total). Each image was taken using 20x objective with 1-μm optical sectioning.

Figure 3. Mosaic Analysis with Double Markers (MADM) model reveals that the loss of IGF1R lead to a 10-fold reduction of GCs.

(A) PCR analysis of purified GCPs/GCs from IGF1R^flox/flox; Cre cerebella shows the presence of a considerable amount of IGF1R flox allele (n=3 mice). (B) qPCR analyses of non-excised IGF1R flox allele in anterior and posterior GCPs/GCs from IGF1R^flox/flox; Cre cerebella shows higher recombination efficiency in anterior than in posterior folia though both are far from complete excision (n=3). (C) Schematic illustration of how MADM generates GFP+, IGF1R-null and RFP+, IGF1R-WT GCPs from a heterozygous, colorless cell through Cre mediated inter-chromosomal recombination. (D-G) Representative images of sagittal sections of P21 cerebellum from MADM-control (D-E) and MADM-IGF1R (F-G) models (inset: zoom-in images of MADM labeled GCs). (H) The comparison of G/R ratio between MADM-control and MADM-IGF1R mice shows that the loss of IGF1R lead to ~10-fold decrease in the G/R ratio (n=6 brains per genotype). (I) The G/R ratio in the anterior (folia II to VI) and posterior (folia VII to X) regions of MADM-IGF1R cerebella at P21 were not statistically different (n=6 brains).

Scale bars: D-G= 1 mm

Statistics in B: student’s t-test; H: One-sample t-test; I: Wilcoxon sign rank test; error bar ± SEM; ns not significant, *p < .05, **p < .01.

Figure 5. The cell cycle progression of IGF1R-null cells was partially rescued upon the inactivation of p27^kip1 in the MADM-IGF1R model.

(A) A hypothetical illustration of the interaction between IGF1 and p27 in GCP cell cycle progression. (B) Schematic illustration of the expected outcomes of IGF1R-WT and null cells when the p27 gene is inactivated. (C) The loss of p27 significantly increased the G/R ratio of GCs in MADM-IGF1R; p27-Null mice at P21 (n=5).

Statistics in C: student’s t-test; error bars ± SEM; ***p < .001.

Figure 4. MADM-IGF1R model reveals IGF1R signaling promotes S phase entry and prevents premature cell cycle exit of GCPs.

(A-B) The spatial separation of the developmental processes from GCPs to GCs in P4 cerebellum. A: Proliferating GCPs labeled with BrdU in the outer EGL (oEGL). B: MADM labeled GCPs/GCs dispersed in distinct layers. (C) Representative image used to quantify GFP+ and RFP+ GCPs in the oEGL of P4 MADM-IGF1R cerebella that are co-stained with BrdU. (D) The percentage of BrdU+ MADM-labeled cells from MADM-IGF1R cerebella at P4 indicated that the loss of IGF1R led to reduced S-phase entry of mutant GCPs in comparison to WT GCPs (n=8). (E) Representative image used to quantify the exit fraction of GFP+ and RFP+ GCPs (reside in iEGL/ML/IGL) of MADM-IGF1R cerebella, e.g. a GFP+ cell in the blue circle; and an RFP+ cell in the white dotted circle. (F) Schematic illustration of two kinds of predicted outcomes: normal vs. premature cell cycle exit of mutant GCPs. (G) The comparison of the exit fraction between GFP+ and RFP+ cells from MADM-IGF1R cerebella at P4 indicated that the loss of IGF1R leads to premature cell cycle exit of IGF1R-null cells in comparison to WT cells (n=8).

Scale bars: A= 30 μm B-E= 50 μm

Statistics in D, G: Wilcoxon sign rank test; error bars ± SEM; ns not significant, **p < .01.

Granule cell density and area analysis

Morphometric analysis on H&E stained images of the mouse cerebellum was conducted using the open-source Fiji image processing software. The area, total cell count, and cell density in the anterior folia was determined via the following methods for each image (10x Magnification): The H&E stained images were converted into a binary mask, the folia area of interest was outlined, and the area occupied by all nuclei (i.e. total nuclei area) was determined by running the Measure command. The average nucleus area of individual nuclei was determined with the Particle Analyzer plugin using min and max area parameters [26–60]. Overall 30–300 nuclei were measured per sample to determine the average nucleus area. The total cell count was calculated by dividing the total nuclei area by the average nucleus size in that area. The cell density in the folia area was calculated by dividing the total folia area by the total cell count in that same area.

To determine the relative number of granule cells per folia, we developed a custom Fiji BeanShell script to measure the area of the IGL, total IGL cell count, and relative fraction of IGL granule cells on H&E stained images. The script input parameters are the H&E stained image after image deconvolution, a Region-of-Interest (ROI) file that defines the IGL outline in the image to be processed, and min and max nuclear area limits. A binary segmentation mask is created for all nuclei in the IGL ROI by applying a local Phansalkar auto-threshold and median filter. Nuclei clusters are split by applying a watershed algorithm to create the final IGL nuclei segmentation masks. The Particle Analyzer is applied sequentially to the total nuclei mask, using the user-specified min and max nuclear area limits, to create ROIs defining all individual nuclei in the input image. The script outputs the count, position, and area for all nuclei and provides a color-coded overlay mask with outlines of the identified nuclei for verification of accuracy.

Statistical analysis

All data represent five or more (in vivo) independent experiments. All quantifications are represented with error bars indicating SEM. One-way ANOVA was used to determine the significance of more than two groups of data in Figures 2A, 3F, and Supplemental Figure 1H with Tukey tests. The Wilcoxon Signed-Rank test was used to determine the significance between pairs of data non-normally distributed in Figures 3I, 4D, 4G, and Supplemental Figure 5C. The student’s t-test was used to determine the significance in Figure 3B, 5C, and Supplemental Figure 3, while the Paired t-test determined the significance in Supplemental Figure 6C. The statistical comparison of the G/R ratios from MADM-Control and MADM-IGF1R (Figure 3H) was determined using a one-sample t-test where the sample mean was compared to 1. All statistical analyses were performed using R and graphs prepared using Prism8 (GraphPad).

Results

IGF1R inactivation in GCPs led to stunted development of the cerebellum

It was previously reported that the conditional knockout (CKO) of the Shh receptor smoothened (Smo) in GCPs resulted in an extremely small cerebellum with abnormal foliation due to massive reduction of GCs (Spassky et al., 2008). Therefore, we hypothesized that, if IGF1 signaling plays a similar role in the development of GCPs, then inactivation of IGF1R in GCPs would phenocopy Smo-CKO, at least partially. To test this hypothesis, we established an IGF1R-CKO mouse model, in which IGF1R-flox with loxP sites in intron 2 and 3 (Dietrich et al., 2000) can be inactivated by Cre transgene under the control of atonal 1 (Atoh1/Math1) gene promoter (Math1-Cre) that is specifically expressed in GCPs (Matei et al., 2005) (Fig. S1A). When we examined phenotypes of IGF1R-CKO mice at weaning age (P21), we found a reduction of cerebellar weight in the IGF1R-null (Δ/Δ) by 19% in comparison to WT (+/+) mice (Fig. 2A). When we examined the gross structure, we found that the reduction in cerebellar size only manifested along the anterior to posterior (A-P) axis, but not the medial to lateral (M-L) axis (Fig. 2B). To clearly assess the cytoarchitecture in the cerebella along the A-P axis, we sectioned the tissue sagittally, and observed stunted development of anterior folia II to VI, but relatively normal development of posterior folia VII to X in Δ/Δ mice in comparison to +/+ and +/Δ mice (Figs. 2C–E).

Similarly, we observed in neonatal age the stunted development of anterior folia II to VI in Δ/Δ mice, which even distorted the single-layered organization of PC cell bodies in the PCL layer (Figs. S1B–G). To evaluate whether the stunted anterior phenotype observed at weaning age is due to the reduced number of granule cells, we measured the area of the IGL which correlates to the number of granule cells and found a significant reduction in the area of anterior folia II to VI in Δ/Δ mice in comparison to +/+, and a milder phenotype in +/Δ mice with the same trend. In contrast, the area of posterior folia VII to X did not significantly differ across genotypes (Figs. 2F; S1H). In conclusion, the loss of IGF1R led to a reduced number of GCs in the anterior but not posterior cerebellum.

Potential contributing factors for the differences in phenotype along the A-P axis

While the conditional mouse model revealed the importance of IGF1 signaling in the development of GCPs, it remains puzzling that there was such a dramatic difference in phenotypes along the A-P axis. We hypothesized three non-mutually exclusive interpretations: technically, Cre mediated excision of IGF1R-flox allele could be more efficient in anterior than posterior GCPs; or biologically, GCPs in the anterior may rely more on IGF1R signaling in comparison to posterior GCPs; or developmentally, compensatory mechanisms from VZ-derived progenitor cells triggered by the loss of GCPs (Wojcinski et al., 2019) could play a more prominent role in the posterior cerebellum.

For the technical hypothesis, we searched the literature on the temporospatial regulation of the Math1 promoter, and found that Math1 promoter turns on early (~E12.5) in the anterior region of the cerebellum but much later in the posterior region (~E15.5) (MacHold and Fishell, 2005). To determine whether the timing difference for Cre expression under the control of the Math1 promoter could lead to different efficiency in IGF1R-flox excision, we first assessed whether non-excised IGF1R-flox allele is present in GCPs/GCs from IGF1R-flox/IGF1R-flox; Math1-Cre mice through PCR. Using the percoll gradient subfractionation method (Lee et al., 2009) after tissue dissociation, we obtained GCPs/GCs at 95% purity then used PCR to assess the presence of IGF1R-flox allele. The positive band of the IGF1R-flox allele in these samples indicated the incompleteness of Cre mediated excision in the GCPs/GCs (Fig. 3A). Next, to determine whether the recombination efficiency is greater in the anterior than in the posterior, we separately purified GCPs/GCs from the anterior and posterior regions of the cerebellum, and then performed q-PCR with genomic DNA to quantitatively compare the level of non-excised IGF1R-flox allele. The lower level of the IGF1R flox allele in anterior than posterior GCPs/GCs indicates that the recombination efficiency is much higher in the anterior folia (Fig. 3B). It should be noted that other literature reports showed that the problem of Math1-Cre excision mainly occurred in folia IX and X (Orvis et al., 2012; Schüller et al., 2008), while the problem of IGF1R-flox excision in this study spanned throughout folia VII to X, probably due to more compacted chromatin structure at IGF1R gene locus than gene loci in other studies. Nevertheless, we conclude that phenotypic differences in the A-P axis could at least be partially explained by more complete IGF1R-flox allele excision in the anterior cerebellum due to longer duration of Math1-Cre activity compared to the posterior region.

While this finding addressed the first hypothesis, the second and the third hypotheses remain possible but cannot be tested with the CKO model due to the uncertainty of the genotype of each cell, and the gross distortion of tissue structure that could trigger compensatory mechanisms, respectively.

MADM, a genetic mosaic model, which circumvents the technical hurdles in the CKO model, was used to investigate the A-P differences in phenotype

To circumvent both problems, we utilized a mouse genetic mosaic model called MADM (Mosaic Analysis with Double Markers) (Zong et al., 2005). Through Cre/LoxP mediated inter-chromosomal mitotic recombination, MADM generates sporadic homozygous mutant cells unequivocally labeled with GFP and their homozygous wildtype sibling cells labeled with RFP from a heterozygous, colorless mouse (Fig. 3C). There are three unique advantages of the MADM system for phenotypic analysis: first, the invariable color-to-genotype matching allows one to directly evaluate functions of a gene of interest in vivo; second, the rarity of homozygous-null cells not only provides single-cell resolution but also completely avoids gross aberrations at the tissue level that could confound phenotypic interpretation; and third, the direct comparison between GFP+ mutant and RFP+ wildtype cells allows the detection of even the subtlest anomaly in mutant cells (Muzumdar et al., 2007); (Espinosa and Luo, 2008); (Gonzalez et al., 2018); (Liu et al., 2011); (Hippenmeyer et al., 2010) (Riccio et al., 2016). To perform MADM-based phenotypic analysis, first we established a MADM-IGF1R model (TG,IGF1R-flox/GT; Math1-Cre) and a MADM-control model (TG/GT; Math1-Cre) (Figs. S2A and B).

While sibling GFP+ and RFP+ cells should have equal developmental potential in MADM-control cerebella hence result in a ratio of 1 between green and red cells (G/R ratio), we predict that there would be fewer GFP+ than RFP+ GCPs (G/R ratio < 1) in MADM-IGF1R mice if IGF1R is critical to promote the development of GCPs. After verifying the normal overall cytoarchitecture of the cerebella in MADM-control and MADM-IGF1R mice (Figs. 3D–G), we quantified the GC expansion by counting GFP+ and RFP+ cells from six P21 cerebella of MADM-control and MADM-IGF1R mice, and calculated the G/R ratio. While G/R ratio in MADM-control mice was ~1, there was a significant reduction in G/R ratio in the MADM-IGF1R model, revealing a crucial role of IGF1R in GCP development (Fig. 3H).

Using this model, we next tested our second hypothesis raised above by asking whether there is a differential reliance on IGF1R signaling between anterior and posterior GCPs. The data showed that the G/R ratio in the anterior (folia II-VI) and posterior (folia VII-X) had no significant difference (Fig. 3I), suggesting that both anterior and posterior GCPs rely on IGF1 signaling to a similar extent. Finally, for the third hypothesis, because of the minimal changes of tissue cytoarchitecture in the MADM model, the observed phenotypes should be cell autonomous and free from the complication of compensatory mechanisms (Figs. 3D–G), Taken together, we conclude that IGF1R is important for GCP development throughout all folia in the cerebellum and that the phenotypic differences along the A-P axis observed in the CKO model is mostly caused by the differences in efficiencies of Cre-mediated excision.

Detailed phenotypic analysis for the role of IGF1R in distinct aspects of GCP development

Next, we investigated more deeply the cause of the reduction of IGF1R-null GCs in MADM-IGF1R mice. We surmised that the P21 phenotype could be due to increased death of mutant GCs in the IGL over time, defective migration of newly differentiated mutant GCs, or reduced expansion of mutant GCPs in the EGL. Increased cell death of mutant GCs was not the cause because the G/R ratio in the IGL remained constant between P21 and P90 (Fig. S3). Defective migration was not the cause because we did not observe an abundance of GFP+ GCs being stuck in the ML at all ages examined. Therefore, we focused our analysis on the role of IGF1R signaling for the expansion of GCPs during early development.

There are three non-exclusive possibilities for reduced GCP expansion: IGF1R-null GCPs undergo cell death resulting in a lower number of progenitors to begin with; IGF1R-null GCPs proliferate less; or IGF1R-null GCPs prematurely exit the cell cycle reducing the number of cell cycles required to produce an optimal number of GCs. We chose P4 cerebella for our analyses since the vast majority of GCPs are actively dividing at this age. First, we assessed cell death of GCPs in the oEGL of P4 cerebellum in MADM-IGF1R mice by staining with cleaved-caspase 3, a marker of apoptosis. While the positive control for cleaved caspase 3 staining worked well, we failed to detect any positive staining in GCPs, either mutant or WT (Fig. S4A,B). Although it is possible that a more sensitive apoptotic marker might detect a few MADM-labeled apoptotic cells, we concluded that subtle changes in cell death would unlikely be a major contributing factor for the dramatic reduction of IGF1R-null cells. Next, to evaluate proliferation of IGF1R-null and WT cells in P4 MADM-IGF1R mice, we injected pups with BrdU 4-hours prior to tissue harvest, and quantified the percentage of BrdU+ GCPs in both IGF1R-null and WT populations in the oEGL where GCPs divide (Fig. 4A and C, Fig S4C). We found that the loss of IGF1R caused a significant reduction in S-phase entry based on reduced incorporation of BrdU in IGF1R-null cells compared to WT cells (Fig. 4D). Lastly, we assessed whether IGF1R-null cells prematurely exit the cell cycle. The spatial separation of cells remaining in the cell cycle (oEGL) versus those that have exited the cell cycle (iEGL,ML,IGL) allows us to compare the rate of cell cycle exit between IGF1R-null and WT cells based on their locations (Fig. 4B and E). Our quantification of the percentage of the “exit fraction” (cells in the iEGL,ML,IGL) among the total number of GFP+ and RFP+ cells in each cerebellum (Fig. 4F, Fig S4C), showed a significantly higher percentage of IGF1R-null cells had exited the cell cycle compared to sibling WT cells (Fig. 4G). Taken together, the loss of IGF1 signaling in GCPs led to both reduced S-phase entry and increased cell cycle exit that in combination result in stunted GCP expansion.

Uncovering the mechanisms of IGF1 that regulates GCP cell cycle progression in concert with Shh

We have shown so far that IGF1R signaling is involved in two critical aspects of GCP cell cycle regulation. However, how IGF signaling acts in concert with Shh signaling needs further investigation. First, we hypothesized that IGF1 signaling could promote cilia resorption to allow cell cycle progression (Fig. S5A), because while cilia are absolutely necessary for Shh signaling to promote cell cycle entry of GCPs (Spassky et al., 2008); (Chang et al., 2019), they must be resorbed for the centrosome at its base to be duplicated for cell cycle to progress. The key rationale for this hypothesis is a report by Yeh et al. which showed that IGF1 promotes the resorption of the primary cilium to permit cell cycle progression in NIH3T3 and retinal pigmented epithelial cells (Yeh et al., 2013). To test this hypothesis, we set out to compare the percentage of ciliated cells between the GFP+ and RFP+ populations. If our hypothesis were correct, we would expect a higher percentage of ciliated cells in the GFP+ population. It should be noted that one technical challenge to studying cilia biology in vivo is the correct assignment of cilia to cells in tissues, especially to an individual GCP in the densely packed EGL. Fortunately, the sparse labeling of mutant and WT cells by MADM ameliorates this problem. We sectioned brains from the MADM-IGF1R model at 10 μm thickness, and then stained brain slices with the widely used ciliary marker Arl13b. To assign cilia to MADM labeled GCPs, a series of digital optical sections (z-stacks) at 0.2mm step size was captured, followed by three-dimensional image reconstructions. Cells displaying a canonical rod-shaped ciliary signal for Alr13b were considered as ciliated, while cells containing a single dot of Arl13b at centrioles were considered non-ciliated. After careful quantification, we were unable to find detectable difference in the percentage of ciliated cells between GFP+ and RFP+ populations (Figs. S5 B–C), and thus concluded that IGF1 signaling might not play a critical role in cilia resorption in GCPs.

Alternatively, we hypothesized that IGF1 could promote cell cycle progression by reducing the level of negative cell cycle regulators, in concert with increased cyclin D expression induced by Shh signaling (Kenney and Rowitch, 2000). It was well known that negative cell cycle regulators such as p27^kip1 accumulate over time to halt S-phase entry and promote cell cycle exit in GCPs (Miyazawa et al., 2000); (Ayrault et al., 2009), and that IGF1 downregulates p27 at both mRNA and protein levels (Mairet-Coello et al., 2009). Therefore, we hypothesize that IGF1 could promote cell cycle progression by keeping p27 at a minimal level (Fig. 5A). When we assessed the level of p27 in GFP+ and RFP+ cells in the oEGL from MADM-IGF1R mice (Fig. S6A and B), we found that the GFP+ population tended to have a higher percentage of cells with a high level of p27 compared to the RFP+ population, in line with our hypothesis (Fig. S6C). To test this hypothesis functionally, we established a compound mouse model, in which p27-flox alleles were bred into our MADM-IGF1R model (Fig. S6D). In this model, p27-flox alleles are excised in all GCPs by Math1-Cre, while IGF1R-WT (RFP+) and IGF1R-null (GFP+) GCPs are generated by MADM sporadically. The resulting GFP+ cells are IGF1R-null and p27-null, and RFP+ cells are IGF1R-WT and p27-null. If the role of IGF1 signaling in GCPs is solely mediated through the downregulation of p27, the removal of p27 in IGF1R-null cells would completely restore their cell cycle progression (Fig. 5B). Phenotypically while the G/R ratio from MADM-IGF1R mice is about 0.1 at P21 (Fig. 3H), we would expect that the G/R ratio of GCs in the MADM-IGF1R; p27-Null model (TG,IGF1R-flox/GT; Math1-Cre; p27-flox) to be significantly greater. Considering the incomplete Cre-mediated excision found in earlier experiments (Fig. 3B), we focused the quantification of GFP+ and RFP+ cells on the anterior cerebella in which excision rate is relatively higher. After the quantification, we calculated the G/R ratio and indeed found a significant increase in the G/R ratio of GCs in the MADM-IGF1R; p27-null model, strongly suggesting that IGF1 signaling-mediated suppression of p27 is an important regulatory mechanism in promoting GCP cell cycle progression (Fig. 5C). Interestingly, the rescue was not complete, i.e. G/R ratio did not return to 1, the possible cause of which will be addressed in detail in the Discussion section.

Changes in the level of IGF1 expression coincides with GCP development temporally

Finally, we asked when and from which cell type is IGF1 produced during cerebellar development. To examine the temporal pattern of IGF1 expression, we performed qRT-PCR using RNA prepared from whole cerebellum at different developmental ages. The level of IGF1 mRNA remained low embryonically, increased slightly at birth, peaked around P4–P7, and then declined quickly at P14 and onward (Fig. 6A). To determine which cell type in the cerebellar environment produces IGF1, we performed in situ hybridization on P5 cerebellum and found that IGF1 was solely expressed in the PCL where PC cell bodies reside, as previously described by Fernandez et al. (Fernandez et al., 2010) (Fig. 6B). A similar pattern with a reduced level of expression of IGF1 was seen in the P10 cerebellum by in situ (Fig. 6C), the same trend detected by qRT-PCR (Fig. 6A). In addition to IGF1, there are two other IGF1R ligands, insulin and IGF2. However, the chance for insulin acting as the key ligand is low because its affinity for IGF1R is 100-fold lower than IGF1 (Schumachers et al., 1991). While IGF2 is expressed in the meninges that could affect GCPs, its affinity to IGF1R is 10-fold lower than IGF1. More importantly, IGF2 is continually expressed into adulthood, while p27 increases and GCPs exit cell cycle, suggesting that it is unlikely the key factor in regulating GCP development (Fig. S8) (Stylianopoulou et al., 1988). We, therefore, conclude that IGF1 is a PC-derived paracrine factor whose expression temporally coincides with proliferative activities of GCs (Fig. 6D).

Figure 6. The expression pattern of IGF1 coincides with the proliferative phase of GCP development.

(A) The expression levels of IGF1 from the whole cerebellum at E15.5, P0, P4, P7, P10, P14, and Adult (P21) quantified by qRT-PCR (n=3 E15.5, n=6 all other ages). Relative quantities were normalized to the level of GAPDH mRNA expression. (B-C) In situ hybridization on sagittal sections of wild type cerebellum from P5 and P10 mice with antisense riboprobes specific for IGF1 showed that it is expressed only in the PCL (inset: zoom-in on PCL). While PCL contains cell bodies of both PCs and Bergmann glia, the diameter of IGF1-expressing cells is ~15 μm, which is similar to the size of PC at this age (Weber and Schachner, 1984) and much larger than Bergmann glia (Bergles et al., 1997; Lippman et al., 2008). Furthermore, the in situ pattern of IGF1 appears to be in a single layer rather than zigzagged pattern of Bergmann glia, suggesting that PC is the source cell type for IGF1. (D) A schematic illustration of how IGF1 expression level correlates with the proliferation and differentiation peaks during GCP development.

Scale bars: B= 800 μm (inset= 20 um) C= 2 mm (inset= 20 um)

Discussion

The present study identifies IGF1 as a key regulator of GCP cell cycle progression during normal mouse cerebellar development. We showed that inactivation of IGF1R signaling in GCPs significantly reduced granule cell numbers at the conclusion of cerebellar development. Furthermore, we demonstrate that the dramatic decrease in GC number is caused by reduced S phase entry and increased cell cycle exit of IGF1R-null GCPs. In addition, our results revealed that IGF1 signaling suppresses the level of p27 to promote GCP cell cycle progression. Finally, we show that Purkinje cells are the source of IGF1, and the expression of IGF1 temporally coincides with the period of GCP proliferation during cerebellar development.

MADM provides single-cell resolution for in vivo phenotypic analysis

Gene knockout is a well-known genetic technique used to investigate the function of a gene in a living organism. To circumvent issues of embryonic lethality and global effects caused by whole-animal knockout, conditional gene knockout (CKO) models were developed to study gene function at the tissue level. This method utilizes tissue specific Cre recombinase to excise a specified DNA sequence flanked by loxP sites. While the CKO model provides spatial and temporal control of gene knockout, there are some technical limitations with this model, including 1) incomplete excision of the gene of interest, especially in a population of fast dividing cells; 2) inconsistent coupling between gene knockout and cell labeling, especially when a low-expressing Cre is used; and 3) a potential for compensatory outgrowth in developing tissues that could mask important phenotypes (Wojcinski et al., 2019).

The MADM system can overcome these issues by design. First, gene knockout is unequivocally correlated to a fluorescent cell label since both events occur through a single Cre/LoxP-mediated inter-chromosomal mitotic recombination event, ensuring that GFP+ cells are always homozygous null for the gene of interest. Second, the simultaneous generation of mutant (GFP+) and wildtype (RFP+) sibling cells provided an internal control to directly compare differences in cellular function, enabling the detection of even the subtlest phenotypes. For example, we observed an anterior specific phenotype in the IGF1R-CKO model could be interpreted as differential reliance on IGF1R signaling in GCPs along the A-P axis. However, through MADM, we confidently nullified this hypothesis since the G/R ratio was the same in both regions of the cerebellum (Fig. 3I). Third, the sparse and specific labeling of mutant and WT sibling cells of MADM provided an opportunity for us to study cilia biology in vivo by overcoming a technical challenge of cilia assignment. Fourth, the sporadic generation of mutant cells by the MADM system minimizes the potential of compensatory outgrowth of neighboring cells when mutant cells die off in a conventional CKO model. While the CKO model only demonstrated a 3.5-fold reduction of GC number in anterior folia, MADM revealed a 10-fold decrease of mutant cells. Fifth, the infrequently labeled cells allowed us to distinguish cell autonomous from non-cell autonomous gene functions. For example, in the CKO model we determined that IGF1R signaling is necessary for GCP cell cycle progression and we observed that the cell body of PCs were distorted. However, in the MADM model, we observed only GCP defects, clearly demonstrating that the clustering phenotype of PCs in the CKO model is a secondary phenotype caused by the loss of a large number of GCs. Finally, the MADM system can be combined with conventional CKO model to study genetic interactions, such as how we investigated the role of p27 in IGF1R signaling (Fig. S5D). Even with all these unique advantages, however, we want to point out that models other than MADM should be used for studies focused on either biochemical analysis of mutant cells or physiological consequences of gene inactivation are the main interest because the small number of mutant cells generated by MADM would be a disadvantage in those situations.

In addition to the utilities of MADM showcased in this study, MADM has also been widely used by many research groups to study developmental processes such as dendritic morphogenesis, neuronal migration, stem cell fate, epithelial cell dispersal, lineage-dependent organization of enteric nervous system, etc (Hippenmeyer et al., 2010); (Espinosa et al., 2009); (Gao et al., 2014); (Packard et al., 2013); (Lasrado et al., 2017). Since MADM-mediated knockout is chromosome-based, with the recent development of MADM system on all 19 autosomes in mice, we anticipate more novel discoveries enabled by this genetic system once it is applied more broadly in the field of developmental biology.

IGF1 signaling positively regulates GCP cell cycle progression through reducing a negative regulator

Using MADM we found that the loss of IGF1R signaling in GCPs hindered S phase entry and increased cell cycle exit. It should be noted that these cell cycle defects are much subtler than the 10-fold decrease of G/R ratio. One potential explanation is that these small changes in cell cycle regulation could cumulatively lead to 3–4 fewer cell divisions, and eventually an order of magnitude decrease of GC number over the period of GCP development, since GCP cell division follow a symmetric pattern (Espinosa and Luo, 2008).

We focused on p27 since it is a well-known negative regulator of cell cycle progression in GCPs and can be downregulated by IGF1 signaling. Our finding suggests that IGF1 and Shh could promote cell cycle progression in concert through regulating two distinct aspects: while Shh promotes the expression of positive regulators such as cyclin D, IGF1 downregulates negative players such as p27. While our rescue experiment identified an interaction between IGF1 and p27, the rescue was not complete. Although inefficient excision of the p27 gene could provide one explanation, IGF1 signaling could also be involved in regulating cell cycle through other mechanisms. First, while our cilia hypothesis produced negative data, it is not fully conclusive because the static method of detecting cilia may not be adequate to capture the dynamic process of ciliogenesis. In the future, a combination of the MADM model with other transgenic reporters such as Fucci mice that label distinct cell cycle phases (Ford et al., 2018) and Arl13b-mCherry mice for cilia monitoring in realtime (Bangs et al., 2015) could provide additional insights. Second, IGF1 was reported to signal through PI3K to upregulate the expression of G₁ and G₁/S phase cyclins, Cyclin D and Cyclin E, respectively (Mairet-Coello et al., 2009) (Kenney and Rowitch, 2000), which would not be restored by p27 knockout in IGF1R-null cells. Third, IGF1 signaling has been found to protect Shh effectors (Gli1 and Nmyc) from proteolytic degradation (Kenney, 2004), which would not be restored by p27 knockout either. Taken together, while our study revealed the role of IGF1 in dampening the level of p27 in GCPs to promote cell cycle progression, additional layers of complexity still merit further investigation.

IGF1 is a PC-derived paracrine factor whose expression correlates with GCP expansion temporally

Cell number scaling has been elegantly studied in the cerebellum: the number of deep nuclei excitatory neurons determines the number of PCs (Willett et al., 2019), and the number of PCs determines the number of GCs (Wetts and Herrup, 1983)(Lewis et al., 2004) and white matter progenitor cells that generate astrocytes and interneurons (Fleming and Chiang, 2015; Fleming et al., 2013). To gain mechanistic insights into cell number scaling, one would need to identify all the paracrine factors involved in this developmental process. While Shh is known to be a PC-derived factor that promotes GCP proliferation, the story is far from complete. For example, one remaining puzzle is how could GCPs promptly exit the cell cycle by the weaning age while the expression of Shh persists into adulthood (Lewis et al., 2004). While some studies have identified inhibitory factors that block GCP proliferation by downregulating key effectors of the Hedgehog signaling pathway (Rios, 2004) (Fogarty et al., 2007) (Anne et al., 2013), our current study added a new layer of complexity attributed to the level of IGF1. We observed that the peak of IGF1 expression coincides with the peak proliferation of GCPs early in development; then, the gradual decline in IGF1 level later in development coincides with increased differentiation of GCPs (Fig. 6D). Therefore, our study suggests that the fast declining of a positive factor IGF1 at the end of GCP development seems to facilitate prompt cell cycle exit of GCPs even though the level of Shh remains high, enabling precise cell number scaling.

Supplementary Material

1

Figure S1. Breeding scheme for the IGF1R-CKO model and additional phenotypic analysis. (A) We established the IGF1R-CKO models by intercrossing two separate stocks; IGF1R-flox and IGF1R-flox; Math1-Cre. This breeding scheme results in +/+, +/Δ, and Δ/Δ mice. (B-G) Representative H&E images of Δ/Δ cerebellum at P5 reveal a distorted PCL, specifically in the anterior region of the cerebellum compared to +/+ cerebellum. Purkinje cells in the posterior region of the cerebellum did not show this phenotype. (H) Granule cell density of anterior folia II to VIab in +/+, +/Δ, and Δ/Δ mice at P21 was not statistically different. Therefore, the area of the IGL is representative of granule cell number in IGF1R-CKO models.

cale bars: = 100 μm

tatistics in H: one-way ANOVA with Tukey test; error bars ± SEM

2

Figure S2. MADM breeding scheme and representative images of MADM labeled GCs in MADM-control and MADM-IGF1R models at P21. (A) We intercrossed two separate stocks to generate the MADM-IGF1R mice, in which GFP-labeled cells are IGF1R-null and RFP-labeled cells are IGF1R-WT. (B) Representative images of MADM-labeled GCs from folia II, V, VIb, VIII, and X from MADM-control (left) and MADM-IGF1R (right) mouse at P21.

cale bars: B= 100 μm

3

Figure S3. The G/R ratio from MADM-IGF1R mice at P21 and P90. The loss of IGF1R did not impact GC survival overtime since the G/R ratio was maintained between these two ages (n=6). Statistics: student’s t-test; error bars ± SEM

4

Figure S4. Representative images of brain sections from P4 MADM-IGF1R mice used to quantify cell death, S-phase entry, and cell cycle exit. (A) Representative images of dying cell in the cortical region of embryonic brain tissue positively stained for cleaved-caspase 3 (CC3) when a critical gene is inactivated (Grey = CC3, Blue = DAPI). (B) Representative image of CC3 staining in P4 cerebellum of MADM-IGF1R mice (inset: zoom-in on a CC3 positive cell). About 100 GFP⁺ cells and 300 RFP⁺ cells per P4 MADM-IGF1R cerebella was counted in the EGL (n=3), and none of them were positive for CC3. (C) Representative images of P4 MADM-IGF1R cerebella sections stained for BrdU (left) and MADM (right) used to quantify S-phase entry and cell cycle exit of IGF1R-WT (RFP+) and IGF1R-null (GFP+) cells (inset: BrdU and MADM co-staining in the oEGL). Yellow line separates the cycling fraction in oEGL from the exit fraction in iEGL/ML/IGL. Scale bars: A-B= 100 μm, C= 50 μm.

5

Figure S5. Testing the hypothesis that IGF1 signaling facilitates the resorption of the primary cilia in GCPs for cell cycle progression. (A) A hypothetical illustration of the role of IGF1 in promoting GCP cell cycle progression through resorbing the primary cilia. (B) Representative image of GFP+ and RFP+ cell in the EGL stained with Arl13b to label cilia. (C) Quantification of the percentage of ciliated cells among all GFP+ and RFP+ cells from MADM-IGF1R mice shows no statistical difference (n=4).

cale bars: B= 200 μm and 20 μm

tatistics in C: Wilcoxon sign rank test; error bar ± SEM; ns not significant.

6

Figure S6. Additional data supporting the p27 hypothesis. (A) Relative intensity of p27 level in the EGL used for semi-quantification (Low level – white circle; Medium level – grey circle; High level – black circle). (B) Representative image of a GFP+ and RFP+ GCP in the outer EGL of MADM-IGF1R cerebella stained for p27. (C) The percentage of p27 level (low, medium, high) in IGF1R-null (GFP+) and IGF1R-WT (RFP+) cells from MADM-IGF1R cerebella at P7 (n=3).

cale bars: A-B= 30 μm

tatistics in C: Paired t-test; error bar ± SEM; *p < .05

7

Figure S7. The breeding scheme for the MADM-IGF1R; p27-CKO model, and representative images of MADM labeled GCs in these models. (A) We established the MADM-IGF1R; p27-CKO mouse model by crossing two separate stocks; TG,IGF1R; p27-flox and GT; p27-flox; Math1-Cre. This breeding scheme results in MADM-IGF1R and MADM-IGF1R-p27-Null mice. (B) Representative images of MADM-labeled GCs from folia II-VIb from MADM-IGF1R (left) and MADM-IGF1R-p27-Null (right) mouse at P21.

cale bars: B= 100 μm

8

Figure S8. Temporal patterns of IGF1 and IGF2 expression based on Allen Brain Atlas. Among all anatomical regions, prepontine hindbrain region in the developing mouse brain where the cerebellum is located was chosen for this plot. © 2008 Allen Institute for Brain Science. Allen Developing Mouse Brain Atlas. Available from: developingmouse.brain-map.org

Highlights.

• IGF1 is a positive regulator for the cell cycle of cerebellar granule cell progenitors (GCPs), which are mostly known to rely on Shh as a mitogenic factor

• MADM system circumvents technical hurdles of the conditional knockout model to reveal the cell autonomous functions of IGF1R in GCP development

• IGF1 promotes S phase entry and reduces precocious cell cycle exit of GCPs

• IGF1 is highly expressed by Purkinje cells early in development when GCPs divide actively, then declines as GCPs commit to cell cycle exit