Mutations in thyroid hormone receptor α1 cause premature neurogenesis and progenitor cell depletion in human cortical development.

Mutations in the thyroid hormone receptor α 1 gene (THRA) have recently been identified as a cause of intellectual deficit in humans. Patients present with structural abnormalities including microencephaly, reduced cerebellar volume and decreased axonal density. Here, we show that directed differentiation of THRA mutant patient-derived induced pluripotent stem cells to forebrain neural progenitors is markedly reduced, but mutant progenitor cells can generate deep and upper cortical layer neurons and form functional neuronal networks. Quantitative lineage tracing shows that THRA mutation-containing progenitor cells exit the cell cycle prematurely, resulting in reduced clonal output. Using a micropatterned chip assay, we find that spatial self-organization of mutation-containing progenitor cells in vitro is impaired, consistent with down-regulated expression of cell-cell adhesion genes. These results reveal that thyroid hormone receptor α1 is required for normal neural progenitor cell proliferation in human cerebral cortical development. They also exemplify quantitative approaches for studying neurodevelopmental disorders using patient-derived cells in vitro.

: THRA mutation-containing iPSCs can be induced to neural fate with variable efficiency (A) Patient-derived iPSCs expressed the transcription factors Nanog, Oct4 and Sox2, as well as the membrane proteins TRA-1- 60  (B) Calcium indicator Oregon Green BAPTA was used as a proxy for action potential firing to measure spontaneous neuronal activity (see Figure 3D). Representative traces show the time course of the fluorescent signal (ΔF/F) observed in the ten most active cells in the field of view.
(C) Representative calcium imaging traces of THRA mutation-containing and control cells at day 50 before and after treatment with the sodium channel blocker tetrodotoxin (TTX) or the AMPA receptor antagonist CNQX. Spontaneous activity returned after washout (traces following washout of CNQX looked equivalent).

RT-PCR
Total RNA from cortical cultures was isolated using Trizol (Sigma), and reverse-transcribed to cDNA using random hexamer primers (Applied Biosciences). Semi-quantitative RT-PCR was performed using primers against FOXG1, PAX6 and GAPDH, and visualized in a Gel Doc XR+ Imager (Biorad).

Western blotting
Protein was extracted from overnight frozen cell pellets at -80°C, using Cell Extraction Buffer (Invitrogen) containing 1x complete mini protease inhibitor (Thermo Scientific

Micropattern chip cultures
Neural progenitor cells were dissociated using Accutase, washed once in N2B27 and strained through a 50 μm cell strainer before counting. 10 6 cells were plated onto laminin-coated CYTOOchips in neural maintenance medium containing 20 ng/ml FGF2. After 24 hours, the medium was replaced with neural maintenance medium without FGF2, and cultures were kept for 1-7 days.

Live imaging
For live imaging, tissue culture dishes containing cells in N2B27 medium were placed in a BioStation CT (Nikon) at 37°C with 7% CO2. Images were acquired in phase and green fluorescence channels at 10x or 20x magnification every 10 minutes for a period of 48-72 hours.

RNA sequencing
For RNA-seq library preparation, total RNA was extracted (as outlined above) from 3 control Gene expression profiles were clustered using the GeneE software, based on Pearson's correlation. Gene ontology (GO) analysis was performed using the PANTHER database (www.panther.org). Enriched GO terms with at least five represented genes and p<0.05 were included in the analysis.

Computational model of human cerebral cortex neurogenesis
The modeling scheme used to analyse the clonal lineage data was based on the findings of a recent in vivo genetic labeling study of cortical neurogenesis in mouse, which showed that cortical radial glia progenitor cells (RGs) transit from a symmetrical proliferative phase to a neurogenic phase, in which they asymmetrically give rise to intermediate progenitor cells (IPCs) with variable but limited neurogenic potential (32). In primates, the progenitor cell compartment is more complex, including ventricular and outer radial glia that interconvert between different subtypes (60,61). While a comprehensive quantitative description of human cortical development was thus not feasible, this analysis aimed to identify robust differences in progenitor cell dynamics between Thrα1 mutant and control cell lines. We considered the evolution of clones in control lines first and, in a second step, compared these findings to the Thrα1 mutant clonal data.

Clonal behaviour in control cultures
As RGs are defined by their long-term self-renewal potential, clones that have lost all Ki67 + cells by 10 days post-mixing (dpm) are assumed to derive from IPCs. The distribution of Ki67clones was consistent with a model in which IPCs cycle at a constant rate and, on each division, self-renew asymmetrically with probability q, or differentiate symmetrically with  fig. S4A). As expected, by 10 dpm, this model predicts that all IPC-derived clones are fully differentiated. From the fraction of Ki67clones at 10 dpm (see Fig. 6A), and assuming that RGs and IPCs are labelled with equal efficiency, it followed that around 25% of cycling cells at day 30, and 31% at day 40, are IPCs.
In mouse neocortical development, RGs transition through a series of symmetric proliferative divisions before entering a phase of asymmetric divisions into IPCs or neurons (32). To determine whether a similar sequence of events could be distinguished in the human data, we considered the joint distribution of Ki67cells, which are mostly or exclusively neurons, and Ki67 + cells, which include RGs and IPCs, in clones (see SI Appendix, fig. S4B). In the D30 data at 6 dpm, some clones contained up to seven Ki67 + and no differentiated cells, indicating that at least a proportion of cycling cells are still proliferating symmetrically. By 10 dpm, however, no clones were observed that consisted of more than seven Ki67 + cells but no neurons.
Therefore, progenitors that initially divided symmetrically are producing neurons by 10 dpm.
At the same time, a similar frequency of seven-cell Ki67 + -only clones was also observed at 10 dpm in the day 40 data. Given the asymmetric division pattern of IPCs, this suggests that RGs do not transition unidirectionally from symmetric to asymmetric divisions.
Instead, we probed whether their dynamics were consistent with a model in which the choice between symmetric and asymmetric divisions is made stochastically at the level of individual RGs. In this model, RGs cycle at a constant rate and, on each division, self-renew symmetrically with probability , asymmetrically produce an IPC with probability , or produce two IPCs with probability 1--. Again, the cell cycle times were taken to follow a Gamma distribution. Since the scale parameter, , significantly affected the outcome in this case, it was included as a parameter to fit.
Fitting the model by weighted least squares to the observed average sizes of 'persisting' clones, meaning clones that retain at least one Ki67 + cell, good accordance was achieved for the D30 control data with = 0.47 ± 0.02 per day, = 0.30 ± 0.05, = 0.65 ± 0.05, and = 1.1 ± 0.3 (Fig. 6B). Importantly, with these parameters, the model predicts the size distribution of 'persisting' clones at 6 dpm and 10 dpm. The total clone size distribution, including fully differentiated clones, was also well predicted (Fig. 6C).
The same parameter choice resulted in a good approximation of the clonal data from control cultures infected with GFP-lentivirus at day 40 (Fig. 6B,C), suggesting that any change in progenitor cell behaviour over this time period is small.

Clonal behaviour in TRα1 mutant cultures
The distribution of Ki67clones in TRα1 cultures was indistinguishable from control cultures, suggesting that IPC behaviour is not affected by TRα1 mutations (see SI Appendix, fig. S4A).
However, the fraction of fully differentiated clones in the D30 data was higher than in controls; only 66% of clones retained Ki67 + cells at 10 dpm (see Fig. 6A). Assuming, as before, that these persisting clones are derived from RGs, the fitting procedure was repeated for the TRα1 mutant data at D30, using the cell cycle parameters and found from the control lines. A good approximation of the average 'persisting' clone sizes was obtained with = 0.15 ± 0.03 and = 0.80 ± 0.05 (Fig. 6B). With these parameters, the model correctly predicts the size distribution of 'persisting' clones and consequently the total clone size distribution (Fig. 6C).
In the D40 clonal data, only 20% of clones contained cycling cells at 10 dpm, and the average clone sizes were markedly decreased (Fig. 6A,B). With = 0.05 ± 0.05 and = 0.10 ± 0.05, the average clone sizes were still well approximated (Fig. 6B); a satisfactory approximation of the clone size distribution was also obtained (Fig. 6C). The fate choice probabilities and are therefore significantly reduced in TRα1 mutant compared to control cultures at D40.

Premature neurogenesis and progenitor depletion
To summarise, the control clonal data at D30 and D40 are well approximated by a highly simplified model of cortical development. In this model, RGs cycle on average once every 51 ± 2 hours. On each division, they choose stochastically between symmetric self-renewal with a probability of 30 ± 5 %, asymmetric division with a probability of 65 ± 5 %, or symmetric differentiation into IPCs. IPCs themselves cycle on average once every 37 ± 3 hours. 65 ± 5 % of IPC divisions are asymmetric and the remainder are symmetric differentiating divisions into two neurons, which results in virtually all IPC-derived clones differentiating fully by 10 dpm.
The TRα1 mutant clonal data is well described by the same model with the same cell cycle parameters, suggesting that the unidirectional lineage hierarchy (RGs producing IPCs which in turn give rise to Ns) and characteristic cellular properties are not affected by the mutations.
Instead, the observed clone sizes and compositions are consistent with a change in RG fate choices upon division. At D30, the probability of symmetric self-renewal of RGs is only 15 ± 3 %, while 80 ± 5 % of divisions are asymmetric. At D40, the vast majority of divisions are symmetric differentiating divisions into two IPCs. Consistently, the estimated proportion of RGs in cultures decreases much faster in TRα1 mutant lines, reaching 20% at day 40.
The cell cycle times estimated from the clonal data agree well with earlier in vitro estimates based on clone sizes at 2 dpm and BrdU incorporation, as well as previously reported results from non-human primates (62). As a further consistency check, dissecting out the Ki67 + cell content of clones, the model provides an independent prediction of the progenitor cell number within clones (see Fig. 6B).
While a more complex model might provide an equally good, or better, description of the data, these results suggest that the simplistic model introduced here contains the minimal necessary rules governing stem cell dynamics in TRα1 mutant and control cortical cultures. Importantly, the dramatic difference in clonal dynamics is largely accounted for by the premature differentiation of RGs into IPCs, without any changes to cell cycle kinetics or lineage hierarchy. the pool of RGs with long-term self-renewal potential is also depleted much earlier than in controls, leading to a reduction in the overall number of neurons produced during cortical development.