Spontaneous thalamic activity modulates the cortical innervation of the primary visual nucleus of the thalamus

Sensory processing relies on the correct development of thalamocortical loops. Visual corticothalamic axons (CTAs) invade the dorsolateral geniculate nucleus (dLGN) of the thalamus in early postnatal mice according to a regulated program that includes activity-dependent mechanisms. Spontaneous retinal activity influences the thalamic incursion of CTAs, yet the perinatal thalamus also generates intrinsic patterns of spontaneous activity whose role in modulating afferent connectivity remains unknown. Here, we found that patterned spontaneous activity in the dLGN contributes to proper spatial and temporal innervation of CTAs. Disrupting patterned spontaneous activity in the dLGN delays corticogeniculate innervation under normal conditions and upon eye enucleation. The delayed innervation was evident throughout the first two postnatal weeks but resumes after eye-opening, suggesting that visual experience is necessary for the homeostatic recovery of corticogeniculate innervation.


Introduction
Visual information flows from the retina to the primary visual cortex (V1) relayed by the primary visual nucleus of the thalamus, the dorsal lateral geniculate nucleus (dLGN).
However, the visual message arriving at V1 contains information provided not only by the retina but also by other brain structures. In the dLGN, visual information from the retina is modulated by signals coming from the superior colliculus, thalamic reticular nucleus, brainstem, and feedback projections from layer 6 (L6) neurons of V1 (Guido, 2018). This corticothalamic loop is one of the largest dLGN inputs (Erişir et al., 1997) and seems to influence spike timing, structure of receptive fields or the amplitude of the response in immature and mature stages (Jurgens et al., 2012;Petrof and Sherman, 2013;Denman and Contreras, 2015;Murata and Colonnese, 2016;Hasse and Briggs, 2017).
Thus, the arrival of relevant information at the visual cortex is likely to depend on the correct assembly and function of the corticothalamic connection between V1 and the dLGN.
The assembly of synaptic contacts between dLGN cells and both retinal and cortical projections follow a tightly regulated spatial and temporal organization. In mice, retinal axons start to innervate the dLGN before birth and, by postnatal day 2 (P2), they cover the entire nucleus (Godement et al., 1984). On the other hand, cortical axons target the ventral edge of the dLGN at P0 but they do not innervate the nucleus during the first postnatal days. Cortical axons start to innervate the dLGN at P3, occupying the entire nucleus by the end of the second postnatal week (Jacobs et al., 2007). Thus, although corticothalamic axons (CTAs) arrive when retinothalamic axons (RTAs) begin to innervate the dLGN, CTAs do not enter the dLGN until RTAs have spread throughout the nucleus.

4
The temporal dynamics of CTA innervation of the dLGN suggests that RTAs could orchestrate this process. Indeed, the absence or premature entry of RTAs into the dLGN accelerates corticogeniculate innervation . However, we still ignore the nature of the mechanism involved in this interaction between retinal and cortical inputs. One possibility is that RTAs secrete or control the secretion of molecular signals that modify the progression of CTAs. For example, the retinal input regulates the synthesis of aggrecan, a repulsive molecule that is present in neonatal dLGN and downregulated at the moment of corticogeniculate innervation (Brooks et al., 2013).
Apart from the trophic effects exerted by molecular signaling, activity-dependent mechanisms could be implicated in regulating innervation in the visual system. For example, CTAs accelerate their entry into the dLGN when retinal activity is disrupted in postnatal mice (Grant et al., 2016). Also, patterns of activity generated in the thalamus might influence corticogeniculate projections. Indeed, we have recently shown that the perinatal thalamus generates an intrinsic pattern of spontaneous activity that is known to regulate the organization of sensory systems (Moreno-Juan et al., 2017;Antón-Bolaños et al., 2019). It is possible that intrinsic patterns of thalamic activity regulate the ingrowth of CTA to dLGN. Here, we found that disrupting the pattern of spontaneous activity delayed the entry of CTAs into the dLGN, a delay that is promptly recovered by visual experience.

Disrupting the pattern of spontaneous activity in the developing visual thalamus
To determine whether innervation of the dLGN by visual CTAs requires spontaneous thalamic activity, we disrupted the activity of the thalamus by overexpressing the inward rectifier potassium channel 2.1 (Kcnj2) in all principal thalamic sensory nuclei. As such, we administered tamoxifen to pregnant Gbx2 CreER ;Rosa26 fs-Kcnj2 mice (referred herein as Th Kir ) on gestational day 10.5. Although Gbx2 is expressed in amacrine cells of the retina at later stages (Kerstein et al., 2020), at this early stage of embryogenesis, the overexpression of Kir2.1 (fused to mCherry) was induced specifically in thalamic neurons ( Figure S1) (Antón-Bolaños et al., 2019). We obtained the activity pattern of the dLGN using brain slices from control and Th Kir mice at P0 loaded with a fluorescent calcium indicator. We subdivided the dLGN area into regions of interest (ROIs) and measured the changes in fluorescence from each ROI ( Figure 1A). The different ROIs became activated asynchronously or as part of synchronous ensembles of different sizes ( Figure 1B-1C). Synchronous calcium events were grouped into two different categories: large propagating ensembles (waves) engaging almost the whole dLGN or relatively small ensembles engaging less than 50% of the nucleus. In control slices, synchronous events encompassed small and large ensembles. In contrast, synchronous calcium events in the Th Kir only included small ensembles ( Figure 1D) and the overall activity was diminished ( Figure 1E). Therefore, and as shown previously for the somatosensory and

Thalamic spontaneous activity is involved in the entry of CTAs into the dLGN
In light of the above, we set out to determine whether an abnormal intrinsic pattern of spontaneous activity in the thalamus might affect the progression of CTAs into the dLGN.
As such, we visualized CTAs in the dLGN by immunostaining against vGlut1 transporter which is expressed at the terminals of corticothalamic neurons or by using the Golli-tau-eGfp mouse that labels L6 and a subpopulation of L5 cortical neurons (Jacobs et al., 2007;Piñon et al., 2009). L5 projections target the lateral posterior nucleus and do not normally arborize within the dLGN. Spontaneous activity did not affect the initial targeting of CTAs since L6 axons were located at the border of the dLGN in both control and Th Kir mice at P0 (Figure 2A). At P4, CTAs had already entered the dLGN of control mice, while they remained at the border in Th Kir mice ( Figure 2B, left panels, and Figure 2C).
Between P6 and P9, CTAs covered more than half the nucleus in control mice, however, Th Kir CTAs covered less than 20% of the nucleus ( Figure 2B, right panels, and Figure   2C). We further confirm the visual origin of the delayed CTAs in the Th Kir model by infecting V1 with a lentivirus expressing Gfp at P3. We processed the brains at P9 and found fewer CTAs in the dLGN of Th Kir than in control mice, as in VGlut1 immunostainings ( Figure 2D). Together these results indicate that the innervation of the dLGN by visual CTAs is delayed when thalamic spontaneous activity is disrupted.
In addition to this role of thalamic activity demonstrated above, corticogeniculate innervation of the dLGN is controlled by retinal input (Brooks et al., 2013;Grant et al., 2016). Retinal projections arborized throughout the dLGN, occupying 90% of its area at P2, when CTAs start to invade the nucleus (Godement et al., 1984;Jacobs et al., 2007;Seabrook et al., 2013). Thus, we tested whether an abnormal ingrowth of RTAs into the dLGN in Th Kir mice might underlie delayed CTAs. To this end, we labeled retinal projections by injecting cholera toxin subunit B into the retina.
We found that contralateral and ipsilateral RTA innervation of the dLGN was normal in Th Kir mice at P2 (Figure 2E-2G). Hence, RTAs targeted and invaded the dLGN independently of thalamic activity and therefore, the delayed invasion of CTAs into the dLGN in Th Kir mice is not due to a defect in RTA innervation.

Disrupting spontaneous activity in the thalamus leads to dLGN neuron death
We observed alterations in dLGN shape when thalamic spontaneous activity was disrupted. While from P0 to P3, the size of the dLGN was comparable to the control, it remained static from P4 onwards, coinciding with the initial phase of CTA innervation ( Figure S2A). To determine whether this lack of growth of the dLGN might be due to an increase in cell-death or the enhanced compaction of dLGN cells, we analyzed caspase-3a activity (Ahern et al., 2013;Lossi et al., 2014) and cell density in the dLGN of control and Th Kir mice. There was a significant increase in caspase-3a immunostaining from P3 to P6 in the dLGN of the Th Kir mice ( Figure S2B and S2C) but no increase in cell density ( Figure S2D and S2E). Hence, the diminished size of the dLGN in Th Kir mice might be due to a higher rate of cell-death, which may in turn be related to the absence of CTAs or to abnormal spontaneous activity (Blanquie et al., 2017).

Retinal inputs modify the spontaneous thalamic activity and regulate the innervation of the dLGN by CTAs
The impact of RTAs on the entry of CTAs into the dLGN is evident upon perturbing the normal development of retinal projections. CTAs enter into the dLGN prematurely when retinal projections are genetically ablated or removed by neonatal enucleation, as well as, when the pattern of retinal activity is abnormal (Brooks et al., 2013;Seabrook et al., 2013;Grant et al., 2016). We corroborated these data using an experimental model (called embBE) where both eyes were enucleated at embryonic day (E)14, before retinal axons arrive at the dLGN ( Figure 3A). Whereas CTAs in control mice recapitulated the previously described developmental invasion of the dLGN, CTAs in embBE mice covered a significantly higher proportion of the nucleus at P2. This acceleration of the dLGN invasion was maintained at least until P7, the last developmental stage studied here ( Figure 3B, 3C, and Figure S3).
It is likely that retinal projections influence the dynamics of corticogeniculate projections via trophic factors and the effect of retinal activity (Brooks et al., 2013).
However, here, we also considered another tentative mechanism. In light of our previous findings, we cannot rule out that the absence of retinal input perturbs spontaneous activity in the dLGN, and that these abnormal patterns may contribute to the defects observed in corticothalamic innervation. To start testing this possibility, we recorded postnatal spontaneous activity in embBE mice using calcium imaging in slices. At P0, dLGN from embBE mice exhibited a higher frequency of large propagating waves compared with control mice (Figure 3D-F). Moreover, while large propagating waves normally vanished by P1 in control littermates, they remained in embBE mice, overlapping with the initial phase of the premature entry of CTAs into the dLGN (Figure 3G and 3H).
Thus, the pattern of spontaneous activity in the dLGN is not normal when retinal input is not present. However, in contrast to the Th Kir model, where the absence of large propagating events was associated with a delayed entry of CTAs, an increased frequency of large propagating events in the embBE model was associated with a premature corticogeniculate innervation.
Next, we wondered what the contribution was of thalamic spontaneous activity to the accelerated entry of CTAs into the dLGN in the absence of retinal input. To address this question, we suppressed large events of synchronous activity from the dLGN of embBE mice by performing bilateral embryonic enucleation in Th Kir mice ( Figure 4A-4C). In this scenario, we found that the CTAs were accelerated but to a lesser extent than in embBE mice (Figure 4D-4F). Together these results corroborated the strong influence of RTAs over CTAs and showed that the temporal dynamics of CTA entry into the dLGN partly relies on its pattern of spontaneous activity. Eye-opening occurs in mice at around P13, an event that is crucial for the final maturation and refinement of the visual system (Rochefort et al., 2009;Ko et al., 2013;Riyahi et al., 2021). As the decreased corticothalamic innervation of the dLGN observed at P9 is rescued by P30, we assessed whether eye-opening might play a role in this recovery. We analyzed the innervation of the dLGN by CTAs in both control and Th Kir mice at P10, P13 and P15, just before and after eye-opening. At P10, a massive invasion of CTAs into control dLGN was evident, while few CTAs were found in Th Kir dLGN ( Figure 5A and 5C). However, after eye-opening, we observed that the profiles of CTA invasion in control and Th Kir mice were indistinguishable ( Figure 5B and 5D).

Visual experience rescues CTA innervation of the dLGN in Th
Visual stimulation through closed eyelids evokes glutamatergic retinal waves and light deprivation in mice from P8 to P14 provokes deficient retinogeniculate segregation (Tiriac et al., 2018), demonstrating that activity driven by light stimulus just prior to eye-opening influences the topographic development of the visual system. Thus, we hypothesized that evoked activity may alter the network properties of dLGN neurons, allowing CTAs to enter the dLGN of Th Kir mice. To test this possibility, we maintained control and Th Kir mice in a dark rearing environment from P5 until P15 before analyzing CTA innervation of the dLGN ( Figure 5E). Dark rearing did not provoke major modifications in CTA innervation in control mice at P15. However, in contrast to the rescue phenotype observed in Th Kir mice reared under normal conditions, there was significantly less CTA invasion of the dLGN in dark reared Th Kir mice after eye-opening ( Figure 5F and 5G). Together these results demonstrate that early visual experience is necessary for the homeostatic recovery of corticogeniculate innervation in juvenile Th Kir mice.

Discussion
Here we describe a novel aspect of the coordination of afferent connectivity to the dLGN by intrinsic thalamic activity. Using a mouse model in which spontaneous activity in the dLGN is disrupted, we have found that CTAs were not able to invade the dLGN in time while RTAs were not affected. The CTA invasion is delayed until eye-opening, when visually evoked activity commences and spontaneous retinal waves subside (Blankenship and Feller, 2010). Visual evoked activity is relevant for this recovery since dark-rearing after eye-opening reduces the delayed CTA innervation to less than 50%.
There are aspects of the development of the corticothalamic connectivity that have been well-characterized, such as its temporal dynamics (including waiting periods), the trajectory of cortical axons throughout the telencephalon and diencephalon, the guidance cues involved, and the specificity between cortical layers and the order of their thalamic targets (Grant et al., 2012;Leyva-Diaz and Lopez-Bendito, 2013). However, other aspects remain elusive, especially those related to the final guidance within the thalamus and the invasion of nuclei.
The neonatal dLGN of mice seems to be a repulsive territory for cortical axons since they wait and accumulate in the ventromedial border of the dLGN until P2-P3 (Jacobs et al., 2007). The extracellular matrix of the dLGN after birth presents high levels of aggrecans, a chondroitin sulfate proteoglycan that normally inhibits axon growth (Brooks et al., 2013). Since L6 neurons express the aggrecan receptor, it is likely that this cue contributes to the neonatal waiting period of cortical axons at the dLGN border. It would be interesting to assess whether the delayed entry of cortical axons into the dLGN observed in Th Kir mice is related to aggrecan levels. If perturbing the activity pattern upregulates aggrecans, or downregulates proteases that cleave aggrecans (Stanton et al., 2011), the dLGN would remain as a repulsive territory for cortical axons beyond the first postnatal days.
Apart from aggrecans, no other permissive or repulsive guidance molecule has been implicated in the invasion of cortical axons into the dLGN. A comprehensive screening of differentially expressed genes in the developing thalamic nuclei is necessary to begin to understand spatial and temporal sorting of the incoming cortical axons. To this end, the analysis should be focused on genes related to guidance, cytoskeleton remodeling, cell-cell interactions, among others. Using this kind of approach, differentially expressed genes have been discovered comparing the dLGN and the primary auditory nucleus of the thalamus at P5. Some of these genes belong to the signaling pathways of netrins, semaphorins, slits, and ephrins (Horng et al., 2009), suggesting that a different set of guidance factors may contribute to the formation of modality-specific connections. In the thalamocortical system, we have previously shown that activity controls the expression of one of these guidance molecules. The developing thalamocortical projections express Robo1 in an activity-dependent manner and, when Robo1 is downregulated upon Kir2.1 overexpression, the growing axons change their speed of advance (Mire et al., 2012).
Thus, it is probable that the activity-dependent modulation of the expression of some, as yet, unknown guidance molecules delays cortical axon entry into the dLGN in the Th Kir mice. However, with the available evidence, it is not possible to discern whether activity has a permissive or instructive role in this process.
From our experiments conducted in embBE mice, it is clear that activity is not the only or even the main factor controlling cortical axon behavior in the developing thalamus. In accordance with previous publications, removing retinal afferences causes premature entry of cortical axons into the dLGN . This early invasion is partially delayed if the pattern of spontaneous activity is disrupted in the dLGN. The interpretation of these experiments is not straightforward because we are manipulating activity in a model where retinal axons are removed, a fact that already causes abnormal activity. But, the following perspective may help us to understand the results. We stress here that similar conclusions regarding the role of activity in CTA invasion can be drawn comparing both embBE versus embBE/Th Kir mice or control versus Th Kir mice. In both cases, we observed that perturbing spontaneous activity delayed the entry of cortical axons, irrespectively of the initial time point of the entry. Thus, we think that these data indicate that there is an activity-dependent contribution.
In any case, our data further supports that a hierarchy exists in the incoming afferences to the dLGN, where retinal axons prevail over cortical axons (Grant et al., 2016). In recent publications, it has been suggested that cortical axons may also modulate initial targeting of the visual thalamus by the RTAs. In these studies, a subpopulation of retinal ganglion cells fails to project to the dLGN in the absence of corticogeniculate innervation (Shanks et al., 2016;Diao et al., 2017). Despite these findings, we showed that the delayed entrance of CTAs did not alter the retinal axons in Th Kir mice. Indeed, retinal axons reached and invaded the dLGN normally. Thus, retinal axons were not affected by the abnormal behavior of cortical axons or by the abnormal pattern of activity in the thalamus.
The delayed corticogeniculate innervation in Th Kir mice resumed at eye-opening suggesting that visual experience somehow forces the cortical axons to enter the dLGN.
Visual experience plays a pivotal role in the maturation of visual neuronal networks, especially in species such as ferrets and cats (Chapman and Stryker, 1993;Crair et al., 1998;Li et al., 2006). In mice, the introduction of vision has a milder effect on circuit function (Hagihara et al., 2015); for instance, it induces the formation of functional subnetworks and it delays, but not precludes, the maturation of visual acuity and the emergence of critical periods (Kang et al., 2013;Ko et al., 2013). To the best of our knowledge, there is no other scenario in which sensory experience shortly after eyeopening drives complete recovery from a major developmental defect. This is a striking result. Although we did not perform functional assays, we observed an innervation pattern that resembles that of control mice. Further research is needed to elucidate the activitydependent mechanisms that drive this recovery and to what extent they have an impact on the modulation of ascending visual information.
Finally, we also found that dLGN growth halted when the cortical axons did not enter the nucleus on time. Previous work has shown that enucleation at P0 leads to a decrease in dLGN size and in the gene expression level of ephrin A5 (Kozanian et al 2015). In the math5 mutant mouse, which represents a model of genetic deafferentation as retinal ganglion cells do not differentiate, the dLGN also shows a decrease in size (El-Danaf et al 2015). However, in this case, dLGN cell density was higher compared with control, suggesting that the reduction in dLGN size does not appear to be a consequence of cell loss. As the size of the dLGN was normal in Th Kir mice until P3, just before the initial 14 entry of CTAs, it is also possible that CTAs could exert a trophic effect on dLGN cells, so the lack of connections between CTAs and dLGN might be triggering their cell-death.
Alternatively, the abnormal pattern of thalamic spontaneous activity might provoke an exacerbated cell-death in the dLGN as has been shown to occur in other brain regions such as the cortex (Blanquie et al., 2017).
In summary, our data indicate that, in addition to retinal input, patterned spontaneous activity in the perinatal thalamus provides a signal that regulates the timing of CTA invasion of the dLGN. These results also provide evidence of an experience-dependent mechanism that helps to recover from early defects in circuits within the visual system.        Scale bar, 100 μm.

Mouse strains
Wild type mice maintained in an ICR/CD-1 background were used for the in vivo experiments, tracing studies and Ca 2+ imaging. The day on which the vaginal plug was detected was designated as E0.5. The Rosa26 flox-stop (fs)-Kcnj2-mCherry mouse line has been described previously (Moreno-Juan et al., 2017). This line was crossed with an inducible thalamic-specific Gbx2 CreER line (Brooks et al., 2013;Seabrook et al., 2013) to generate Gbx2 CreER/+ ;Rosa26 fs-Kcnj2 double mutant embryos, referred to as Th Kir . Tamoxifen induction of Cre recombination in the double mutant embryos was performed by gavage administration of tamoxifen (5 mg dissolved in corn oil, Sigma) at E10.5 to specifically target the principal sensory thalamic nucleus. The administration of tamoxifen to pregnant mice hinders vaginal delivery. To avoid abortion, we administrated 125 mg/Kg of progesterone (depo-progevera) intraperitoneally (i.p.) at gestational day 14.5 to delay delivery and mice were born via C-section at E19.5. To guarantee safe nursing, we placed newborns with foster mothers. For Th Kir experiments, CreERT2-negative littermates were used as controls. In the Golli-tau-Gfp mouse line, the golli-promoter of the golli-myelin basic protein gene drives reporter gene expression (enhanced green fluorescent protein (eGFP) fused to bovine tau microtubule protein) selectively in L6 and a subpopulation of L5 cortical neurons, as previously described (Jacobs, 2007). We crossed this line with the Th Kir animals. Mutants of either sex were analyzed at stages from P0 to P30.

Histology
For immunohistochemistry, mice were perfused with 4% paraformaldehyde (PFA) in PBS (0.01M), and their brains were removed and post-fixed overnight in the same fixative. For fixation of the retinas, the eyes were dissected out and fixed with 4% PFA for 30 minutes. Immunohistochemistry for all the antibodies was performed on 80-μm sections cut using a vibratome.
Brain sections were first incubated for 1h at room temperature (RT) in a blocking solution containing 1% BSA (Sigma), 2% Donkey Serum (Biowest) and 0.25% Triton X-100 Finally, the sections were counterstained with the fluorescent nuclear dye DAPI (Sigma-Aldrich). For Nissl staining, the 80-μm sections were mounted on slides and dried overnight. The slides were incubated in a cresyl violet solution for 30 seconds to 2 minutes, dehydrated in an ethanol series (70% ethanol for 3 minutes, 70% ethanol for 3 minutes, 90% ethanol for 3 minutes, 90% ethanol for 3 minutes, 100% ethanol for 3 minutes), passed through Xylol (2 x 5 minutes) and immediately mounted with Eukitt® (ThermoFisher).

Viral tracing of visual corticothalamic axons
For cortical axonal tracing at postnatal day 9, a lentivirus containing Gfp (pLM-CMV-tauEGFP) was injected into the V1 cortex of control and Th Kir mice at P3. Pups were anesthetized under hypothermia and 5 pulses of 69 nL were injected into the V1 with a nanoliter injector (WPI). Four days later, the animals were perfused with 4% PFA in PBS (0.01M) and their brain was dissected out and post-fixed overnight in the same fixative solution. Vibratome sections (thickness: 80 μm) were counterstained with DAPI (Sigma-Aldrich).

Tracing of retinal axons
For axonal tracing at postnatal stages, P0 pups were anesthetized under hypothermia and P6 pups were anesthetized with isoflurane. An incision was made over the eyelid of one eye and it was injected with CTB conjugated to Alexa-647 (Mol. Probes C34778) using a nanoliter injector (WPI). The animals were perfused with 4% PFA in PBS (0.01M) 2, 3, or 7 days later, and their brain was dissected out and post-fixed overnight in the same fixative. Vibratome sections (thickness: 80 μm) were then counterstained with the fluorescent nuclear dye DAPI (Sigma-Aldrich).

In utero enucleation
For in utero enucleation, wild type or Th Kir pregnant females (gestational day 14.5) were deeply anesthetized with isoflurane to perform laparotomies. The embryos were exposed and both eyes were cauterized in half of the litter. The surgical incision was closed and the embryos were allowed to develop until postnatal stages.

Calcium imaging in slices
The brains of postnatal mice (P0 and P1) obtained after they were sacrificed by decapitation were used for calcium imaging experiments. The brains were immediately dissected out and 300-μm thick coronal slices were cut using a vibratome (VT1200 Leica Microsystems Germany) using a slicing solution bubbled with 95% O2 and 5% CO2 and containing (mM): 2.5 KCl, 7 MgSO4, 0.5 CaCl2, 1 NaH2PO4, 26 NaHCO3, 11 glucose and 228 sucrose. The slices were kept at RT in oxygenated artificial CSF (aCSF) containing (mM): 119 NaCl, 5 KCl, 1.3 MgSO4, 2.4 CaCl2, 1 NaH2PO4, 11 glucose. To load the calcium indicator, the slices were incubated at 37 ºC for 30 minutes in oxygenated aCSF containing 5 μl of Cal520 AM (AAT Bioquest), 1 mM DMSO and 20% pluronic acid. The slices were then placed under a Leica DMi8 microscope in a recording chamber perfused with oxygenated aCSF at 34 ºC. We acquired images every 300 ms for 15 minutes using a Hamamatsu ORCA-Flash 4.OLT camera. A minimum of three timelapse movies were obtained per slice. We used one slice per animal and each data point in the graph corresponds to one independent observation. Calcium imaging data were analyzed as extensively described elsewhere (Moreno-Juan et al., 2017;Antón-Bolaños et al., 2019). Briefly, the dLGN was delineated in each slice and this area was parcellated into ROIs using a grid of 6x6 pixel squares. Each ROI had the approximate size of a cell.
Calcium events were detected in each ROI and the fraction of the total ROIs active was quantified throughout the movie.

Dark rearing experiments
For the DR experiments, Th Kir pups and their control littermates were housed into a completely dark environment with their lactating dam from P5 to P15. After this, the pups were perfused with 4% PFA in PBS (0.01M). Their brain was dissected out and postfixed overnight in the same fixative. Vibratome sections (thickness: 80 μm) were then subjected to immunohistochemistry.

Data analysis and statistics
The analysis of the calcium imaging experiments was described in detail elsewhere (Moreno-Juan et al., 2017;Antón-Bolaños et al., 2019). In order to delineate the dLGN we excluded the optic tract and we followed the clear separation of the dLGN with the ventral lateral geniculate nucleus and the ventral posterior medial nucleus. ImageJ software was used to quantify CTA and RTA invasion of the dLGN. Coronal slices (thickness: 80 µm) were obtained from the animals subjected to the different conditions and immunostained against vGlut1 or GFP. At least three dLGN sections were analyzed from each animal, correcting for the fluorescence background. Threshold fluorescence images were analyzed using ImageJ, analyzing the particles of the pixels that cover the dLGN and measuring the proportion of pixels occupied by CTAs. For the analysis, we used at least three slices from each animal calculating the percentage of innervation. The average value of the three sections was considered an independent observation.
Caspase3a positive cells were quantified in coronal sections (thickness: 80 μm) and normalized to the area of each dLGN. The whole dLGN was measured from P3, P4 and P6 control and Th Kir mice, and the results normalized to the mean of the controls, representing the fold-change in Caspase3a positive cells in the Th Kir mice.
Statistical analysis was carried out with GraphPad Prism6 TM and Matlab TM , and the data are presented as mean and SEM. Statistical comparison between the groups was performed using an unpaired two-tailed Student's t test and when the data failed a Kolmogorov-Smirnov or a Shapiro Wilk test of normality, a two-tailed, non-parametric Mann-Whitney U-test was used. A one-way ANOVA test was performed on the embryonic enucleation in Th Kir mice experiments. A two-way ANOVA test was performed on the embryonic enucleation in Th Kir mice and for the DR experiments. No statistical methods were used to predetermine the sample size, although our sample sizes are considered adequate for the experiments and consistent with the literature (Brooks et al., 2013;Seabrook et al., 2013;Grant et al., 2016). The allocation of mice to the experimental groups was not randomized.