Aberrant development of excitatory circuits to inhibitory neurons in the primary visual cortex after neonatal binocular enucleation

The development of GABAergic interneurons is important for the functional maturation of cortical circuits. After migrating into the cortex, GABAergic interneurons start to receive glutamatergic connections from cortical excitatory neurons and thus gradually become integrated into cortical circuits. These glutamatergic connections are mediated by glutamate receptors including AMPA and NMDA receptors and the ratio of AMPA to NMDA receptors decreases during development. Since previous studies have shown that retinal input can regulate the early development of connections along the visual pathway, we investigated if the maturation of glutamatergic inputs to GABAergic interneurons in the visual cortex requires retinal input. We mapped the spatial pattern of glutamatergic connections to layer 4 (L4) GABAergic interneurons in mouse visual cortex at around postnatal day (P) 16 by laser-scanning photostimulation and investigated the effect of binocular enucleations at P1/P2 on these patterns. Gad2-positive interneurons in enucleated animals showed an increased fraction of AMPAR-mediated input from L2/3 and a decreased fraction of input from L5/6. Parvalbumin-expressing (PV) interneurons showed similar changes in relative connectivity. NMDAR-only input was largely unchanged by enucleation. Our results show that retinal input sculpts the integration of interneurons into V1 circuits and suggest that the development of AMPAR- and NMDAR-only connections might be regulated differently.

Electrophysiology. Coronal slices (400 µm) from control and enucleated animals at P16/P17 were used for all recordings. Neurons in the binocular region of the V1 were selected for recordings. For each experimental group, at least three animals were used for making brain slices. Typically, we recorded from two brain slices from each animal. Slices were cut with a vibratome (VT1200, Leica) in ice cold modified artificial cerebrospinal fluid (ACSF) containing (in mM): 212.7 sucrose, 2.6 KCl, 1.23 NaH 2 PO 4 , 26 NaHCO 3 , 10 glucose, 3 MgCl 2 , 1 CaCl 2 (pH 7. 35-7.4). Slices were incubated at 30 °C for 30 min, then kept at room temperature in ACSF containing (in mM): 130 NaCl, 3 KCl, 1.25 NaHCO 3 , 10 glucose, 1.3 MgSO 4 and 2.5 CaCl 2 (pH 7. 35-7.4). All ACSF were equilibrated with 95% O 2 -5% CO 2 . Fluorescence-targeted recordings were performed as previously 23 ; borosilicate recording electrodes (4-9 MΩ) were filled with internal solution containing (in mM): 115 cesium methanesulfonate, 5 NaF, 10 EGTA, 10 HEPES, 9 CsCl, 3.5 MgATP, 0.3 NaGTP and 3 QX-314 (pH 7.25; 300 mOsm); 0.5% Biocytin was added to the internal solution. Red fluorescent protein-expressing interneurons in L4 of the binocular region of V1 were identified with epifluorescence microscopy. Recordings were performed at room temperature in high-divalent ACSF containing (in mM): 124 NaCl, 5 KCl, 1.23 NaH 2 PO 4 , 26 NaHCO 3 , 10 glucose, 4 MgCl 2 and 4 CaCl 2 . Picrotoxin (100 μM) was added to the bath solution during recordings to block GABA A receptor-mediated currents 23 . All drugs and chemicals were purchased from Sigma-Aldrich unless specified otherwise. Data were acquired with a voltage-clamp amplifier (Multiclamp 700B; Molecular Devices) and a DAQ board (NI PCI-6259, National Instruments) controlled by EPHUS software 33 running in MATLAB (The Mathworks). Membrane potential was corrected with 10 mV of estimated liquid junction potential. Series resistance was typically between 20 and 40 MΩ. Input resistance was similar between control and enucleation conditions. Laser-scanning photostimulation (LSPS). LSPS was performed as described previously 23,34 . Highdivalent ACSF containing 0.8 mM caged glutamate [N-(6-nitro-7-coumarylmethyl)-l-glutamate] 35,36 was used as the bath solution during recording. UV laser (3510-100, DPSS Lasers Inc.) stimulation (355 nm, 1-ms pulses) was delivered through a 10 × water immersion objective (Olympus UMPLFLN10XW). Laser-scanning directions were controlled by a mirror galvanometer (6210H, Cambridge Technology). Laser power on the specimen was between 22 and 25 mW. For each map, a rectangular array of up to 30 × 30 sites with 40 μm spacing was stimulated once at 1 Hz in a pseudorandom order. This stimulation paradigm evokes action potentials in neurons at the stimulation sites with similar spatial resolution (~ 100 μm) in all cortical layers between the control and enucleation groups. Putative monosynaptic currents (EPSCs) in GABAergic interneurons were classified by the latency of the evoked currents. We interpreted EPSCs with latencies less than 10 ms as the results of direct activation of the glutamate receptors on the patched cell. EPSCs with latencies between 10 and 50 ms are interpreted as monosynaptic EPSCs ( Fig. 2A). A minimal peak amplitude of 10 pA was used to minimize false positive EPSCs. The peak amplitude and the transferred charge (the area under the EPSC trace between 10 and 50 ms) of the EPSCs were used as measures of connection strength. Based on previous experiments using TTX, we estimate that 90% of the classified monosynaptic EPSC events were correctly identified using these event criteria 23,34 . For each neuron, recordings were made at − 70 mV and + 40 mV to map, respectively, AMPAR-mediated inputs and www.nature.com/scientificreports/ to include NMDAR-mediated current for each stimulation site. Input maps obtained at + 40 mV and − 70 mV holding potential were compared for each cell: at any spatial location, an input that was present only at + 40 mV was identified as NMDAR-only input (Fig. 4A,B). A given presynaptic neuron may make multiple synapses with a postsynaptic cell and only some of them might be NMDAR-only synapses. Our method cannot resolve such a scenario and would treat this connection as AMPAR-mediated. Thus, on the individual synaptic level we may underestimate the number of NMDAR-only synapses. It is possible that binocular enucleation could have introduced broad excitability changes in neurons that would have influenced the efficacy with which uncaged glutamate would evoke action potentials, subsequently changing the spatial resolution of our stimulation method. To examine this, we used loose-patch recordings to record action potentials evoked by photostimulation 37 (Fig. 1). Patch pipettes were filled with filtered highdivalent ACSF. Typically, 1-2 action potentials were reliably evoked by our stimulation from sites within 150 μm of the cell body (Fig. 1D,E). Data analysis and statistics. Data were analyzed in MATLAB with custom routines. Features in the bright-field image were used to identify cortical layer boundaries as previously described 34,38,39 . Monosynaptically-evoked EPSCs were distinguished from directly-evoked EPSCs by a latency threshold as detailed above. We then quantified the number of stimulation locations that give rise to synaptic EPSCs and divided them based on their layer of origin. The fractional input from each layer group was the number of stimulation locations that could evoke synaptic EPSCs from each layer group divided by the total number of stimulation locations that could evoke synaptic EPSCs from all layers. To reveal the general pattern of the input locations, we aligned input maps with the soma of the recorded neuron and calculated the fraction of cells that received an input from each stimulation location (Fig. 2B) and the average EPSC strength. The locations close to the soma and dendrites of the recorded cells were not included in the analysis due to direct activation (Fig. 4B). Therefore, the average input  Input maps are aligned to the soma and averaged to generate average input maps (right). (C) Average spatial connection probability maps of AMPAR-mediated inputs to L4 Gad2 interneurons in V1 at P16/P17 in control and enucleated animals (enucleated at P1/P2) (n = 14 control and n = 12 enucleated). Maps show the fraction of cells that received an input from a particular spatial location. AMPAR-mediated inputs from subgranular layers are decreased after enucleation. The length of the scale bars marking the layer boundaries is 100 μm.  Changes in the input locations can also lead to changes in the average input location map even without significant changes in the input area. Inputs consistently from the same location for all cells will increase the probability of finding presynaptic input in that location on the average input location map. Inconsistent input locations will lower the probability of finding presynaptic inputs in the average input location map. This reflects how consistent the spatial distribution of the inputs is from cell to cell. To quantify the consistency of the spatial distribution of the inputs, for each cell, we calculated the distribution of the input locations in 80 μm bins along the medial to lateral axis in each layer group. For each layer group, we use the distributions from any combination of 2 cells in the same experimental group to calculated Pearson's Linear Correlation Coefficient as a measure of input location variability (Fig. 3A). Consistent input locations will generate similar distribution and the pairwise correlation coefficients were closer to 1. Pairwise correlation coefficients were subsequently compared between the control and enucleated conditions.
For group comparisons we first used a Shapiro-Wilks test to evaluate if data were normally distributed. If both groups were normally distributed, a two-sample t-test was used to test the significance of the difference. Otherwise, a Wilcoxon rank-sum test was used. Significant differences were marked as (*) P < 0.05 or (**) P < 0.01. Comparisons with P > 0.05 were not marked in the figures. Median and interquartile range (IQR) were reported for each data group. Cohen's d and r were used to quantify the effect size for two-sample t-test and Wilcoxon rank-sum test, respectively. Cohen's d values larger than 0.8 indicates large effect. r values larger than 0.5 indicates large effect. Descriptive statistics and statistical test methods were detailed in the main text and the tables.

Results
We sought to determine if retinal input is required for the normal maturation of glutamatergic connections received by layer 4 (L4) GABAergic interneurons in V1. We thus performed bilateral enucleation in mouse pups at P1/P2 when GABAergic interneurons receive few glutamatergic connections from cortical excitatory neurons 21 . As cortical firing patterns in V1 mature at ~ P16/17 29 just before the onset of the critical period 40-42 , we assessed the effect of neonatal enucleation at P16/P17. The cortical sources of glutamatergic connections to L4 GABAergic interneurons were mapped by laser-scanning photostimulation (LSPS) using caged glutamate and whole-cell patch clamp recordings in acute brain slices in vitro 23,34 .
We first tested if enucleation changed cortical neurons' responses to glutamate uncaging, thereby changing the spatial resolution and efficacy of our LSPS technique. Thus, we performed cell-attached recordings from excitatory neurons in V1 from all cortical layers (control: L2/3/4, n = 22 cells, L5/6 and SP, n = 22 cells; enucleation: L2/3/4, n = 22 cells, L5/6 and SP, n = 25 cells) during glutamate photostimulation (Fig. 1B). We found that enucleation did not alter the total number of stimulus locations from which action potentials could be evoked ( Fig. 1C; Table 1), the distance of stimulation locations where action potentials can be evoked ( Fig. 1D; Table 1), the average number of evoked action potentials ( Fig. 1E; Table 1), nor the average latency of the evoked action potentials ( Fig. 1F; Table 1). Thus, the excitation pattern of the excitatory neurons in response to photostimulation and the spatial resolution of LSPS in V1 were comparable between control and enucleated animals in our experimental paradigm.
Proportion of AMPAR-mediated inputs from subgranular layers decreased in Gad2 interneurons after enucleation. We compared the source of glutamatergic inputs to L4 GABAergic interneurons in V1 in slices at P16/17 from control and enucleated animals. To perform targeted recordings from GABAergic interneurons, we used the offspring of Gad2-Cre (a pan-interneuron marker) 30,43 and floxed-tdTomato transgenic mice (Ai9) to drive red fluorescent protein expression in Gad2-positive GABAergic interneurons ( Fig. 2A). For each recorded GABAergic interneuron, we measured AMPAR-mediated EPSCs at a holding potential of -70 mV while using LSPS glutamate uncaging to stimulate cortical neurons ( Fig. 2A). The stimulation grid covered a cortical area of 0.8 × 1 mm 2 around the recorded neuron, encompassing all cortical layers (Fig. 1B). For each recorded neuron, we generated an input map showing the locations of the presynaptic inputs by using the existence of monosynaptic EPSCs as an indicator of presynaptic inputs from the stimulation locations (Fig. 2B). Then we aligned the somata of all recorded neurons under each condition and averaged the input location maps to generate a 2-dimensional map reflecting the spatial probability of connections in each recorded population ( Fig. 2B,C). By averaging the input location maps along the dorsal to ventral dimension we generated a laminar input probability plot (Fig. 2D). Qualitative comparison of both plots between control and enucleated animals suggested that cells in enucleated animals had fewer AMPAR-mediated inputs from subgranular layers L5/6 ( Fig. 2C,D). To confirm these qualitative observations, we next performed analysis on a single cell basis. We quantified the total number of locations of AMPAR-mediated inputs to each cell by summing the stimulation sites that resulted in an evoked EPSC (Fig. 2E, left). We found that there was a trend toward a decreased total number of AMPAR-mediated input locations (control median: 48, IQR: 31, enucleation median: 26, IQR: 25, P = 0.117, effect size: r = 0.31, Wilcoxon rank-sum test). Given that the average and laminar maps (Fig. 2C,D) suggested that differences were largest for L5/6, we next performed a laminar analysis. Separately comparing the number of input locations from each layer revealed that there were fewer AMPAR-mediated input locations from L5/6 in cells from enucleated animals than in control (Fig. 2E, right; Table 2). Accordingly, the fraction of AMPAR-mediated input locations from L5/6 was lower in cells from enucleated animals than in control, while the fraction of inputs from L2/3 was increased in cells from enucleated animals ( Fig. 2F; Table 2). In contrast, the total number of AMPAR-mediated input locations originating from L2/3 and L4 was not altered by enucleation.
Besides causing a reduced number of input locations from L5/6, enucleation could subtly alter the spatial pattern of AMPAR-mediated input locations. In particular, while the probability of presynaptic cell location in the average maps suggested differences in L2-4, we did not find a difference in the number of AMPAR-mediated input locations from L2-4. Since heterogeneity in the spatial pattern between cells could lead to reduced probability of presynaptic cell location in the averaged map, we next investigated if the locations of the presynaptic connections were more variable between the control and enucleation condition. To quantify such spatial changes, we calculated the marginal distribution of the AMPAR-mediated connections along the dorsal-ventral axis (thus parallel to the pia) for cells in each layer group and then tested how similar this marginal distribution was within each group (Fig. 3A). To quantify similarity, we computed the pairwise correlation between cells within the control and enucleation group (Fig. 3A). Higher pairwise correlations indicate that the marginal distributions of the connections along the dorsal-ventral (intralaminar) axis are more similar between the cells. Such similarity would give rise to a higher connection probability and more color-intense area on the average input location probability maps. This analysis showed that cells from enucleated animals showed lower correlations of the input locations in all cortical layers than cells from control animals ( Fig. 3B; Table 3). These results suggest an increase in heterogeneity in the intracortical AMPAR-mediated circuits impinging on L4 GABAergic interneurons after www.nature.com/scientificreports/ enucleation. Thus, our data suggest that normal maturation and spatial refinement of AMPAR-mediated inputs to GABAergic interneurons in L4 of V1 requires peripheral input.

NMDAR-only connections are not altered in Gad2 interneurons after enucleation. While
AMPAR-containing synapses represent the mature state of glutamatergic synapses, developing cortical circuits also have NMDAR-only synapses, which is a feature of the developing nervous system [44][45][46] . The fraction of NMDAR-only input to cortical GABAergic interneurons is higher in the first postnatal week than at later ages 23 .
Since NMDAR-only inputs can be gradually converted into AMPAR-containing inputs by neural activity during postnatal development 24,45,47 , we tested if enucleation altered the developmental reduction of the NMDAR-only inputs and if such changes mirror the observed changes in the AMPAR-mediated inputs. For each neuron, we performed recordings at holding potentials of − 70 mV and 40 mV during LSPS glutamate uncaging (Fig. 4A). We identified the spatial locations of NMDAR-only inputs as stimulation sites where monosynaptic EPSCs were observed only at 40 mV holding potential but not at − 70 mV (Fig. 4B). For each recorded Gad2 interneuron we generated an NMDAR-only input location map and then aligned the maps of all cells in each condition at the soma and averaged them, resulting in an input location probability map for NMDAR-only inputs (Fig. 4C). Qualitative inspection showed that few consistent NMDAR-only inputs were present but that the spatial pattern of NMDAR-only inputs to Gad2 interneurons was similar between control and enucleated animals. The laminar plots revealed a high variance of input pattern from the columnar profile of each cell (Fig. 4D). Quantifying the total number of NMDAR-only inputs showed a similar amount of input between enucleated and control animals (Fig. 4E, left; control median: 6, IQR: 8, enucleation median: 5, IQR: 13, P = 0.817, effect size: r = 0.05, Wilcoxon rank-sum test). Quantification of the laminar sources of inputs revealed no obvious differences between cells from enucleated and control animals (Fig. 4E, right; Table 4), but small increases were present in the average proportion of inputs from L4 ( Fig. 4F; Table 4). Analysis of the correlations of the input locations of NMDAR-only inputs revealed that the location of NMDAR-only inputs from L4 was less variable in the enucleated animals ( Fig. 4G; Table 4).
We next computed the overall percentage of NMDAR-only input (NMDAR-only input to AMPAR-mediated input ratio) and found that it was comparable between control and enucleation group (Fig. 4H, left; control median: 0.16, IQR: 0.16, enucleation median: 0.2, IQR: 0.53, P = 0.395, effect size: r = 0.17, Wilcoxon rank-sum test). Separately computing the percentage of NMDAR-only input in different layer groups showed that the laminar distribution of NMDAR-only input was comparable between control and enucleation group (Fig. 4H, right; Table 4).
Together our findings showed that the absence of retinal input did not alter NMDAR-only inputs to L4 Gad2 interneurons in V1 identified by our method. Since we find that enucleation prevents the development of AMPAR-mediated connections, these results suggest that the increase in AMPAR-mediated connections and decrease in NMDAR-only connections might be independently regulated.  Table 3. Statistics in Fig. 3. IQR: interquartile range.   Cohen's d = 0.001, two-sample t-test) was similar between conditions. The average EPSC transferred charge in separate layers was also similar between control and enucleation animals (Fig. 5B, right; Table 5). The average spatial map of the transferred charge of EPSCs from NMDAR-only connections (Fig. 5C) was also similar between conditions. Enucleation did not change the overall transferred charge of NMDAR-only connections (Fig. 5D, left; control median: 0.24 pC, IQR: 0.12 pC, enucleation median: 0.18 pC, IQR: 0.18 pC, P = 0.339, effect size: r = 0.2, Wilcoxon rank-sum test) nor the average EPSC transferred charge from different layers (Fig. 5D, right; Table 5). Together, our results suggest that the synaptic strength of glutamatergic connections received by V1 Gad2 interneurons is not affected by neonatal enucleation.

Proportion of AMPAR-mediated inputs from subgranular layers decreased in PV interneurons after enucleation. GABAergic interneurons labeled by
Gad2-Cre comprise different cell types 30,43 , which may have different developmental trajectory [48][49][50] and may respond differently to removal of sensory activity 51 . The results that we observed from Gad2 interneurons could represent the average of differential changes in different types of GABAergic interneurons. Accordingly, we used a more restricted mouse line-PV-Cre transgenic mouse, to selectively probe changes in fast-spiking, parvalbumin-positive, interneurons 52 in L4 of V1. PV interneurons play a key role in gating the critical period 25 , thus factors affecting their maturation can influence the later critical period. We thus repeated the above studies in this mouse line and compared AMPAR-mediated and NMDAR-only connections onto PV interneurons (Fig. 6A) at P16/P17 in control and neonatally enucleated mice (control: n = 12 cells, enucleation: n = 13 cells).
Qualitative inspection of the average connection probability maps (Fig. 6B) and laminar input plots (Fig. 6C) showed an increase of AMPAR-mediated inputs from L2-4 and a decrease of AMPAR-mediated inputs from L5/6. Therefore, we quantified the total number of AMPAR-mediate input locations in each PV interneuron. The total number of input locations across all layers was similar between control and enucleation condition (Fig. 6D, left; control median: 59, IQR: 56, enucleation median: 88, IQR: 42, P = 0.181, effect size: Cohen's d = 0.55, two-sample t-test). Since Gad2 positive interneurons showed laminar specific differences in AMPAR-mediated inputs after enucleation, we performed a laminar analysis. Separately comparing the number of AMPAR-mediated inputs from each layer revealed that PV interneurons from enucleated animals had an increase of AMPAR-mediated Table 4. Statistics in Fig. 4. IQR interquartile range.  www.nature.com/scientificreports/ inputs from L2/3 and L4 than PV interneurons from control animals (Fig. 6D, right; Table 6). Accordingly, the average proportion of AMAPAR-mediated inputs from L2-4 to PV interneurons was larger in enucleated animals than in control and the fraction of inputs from L5/6 was reduced in enucleated animals ( Fig. 6E; Table 6). Besides increased connectivity from L2-4 in PV cells from enucleated mice, the average connection probability maps had also suggested a difference in the input locations from L5/6. Since the total numbers of connections from L5/6 were similar between cells from enucleated and control mice, such differences in the average maps  Table 5. www.nature.com/scientificreports/ could be caused by changes in the spatial origins of the inputs. Similar to our approach above, we calculated the correlation of input locations for each layer group. We found that the correlation of input locations was lower for inputs from L5/6 in PV interneurons from enucleated mice ( Fig. 6F; Table 6). This larger heterogeneity of L5/6 inputs likely resulted in the observed lower average connection probability from L5/6 in enucleated mice. In contrast to the lower correlation for L5/6 connections we found larger spatial correlation of input originating in L4, suggesting that the locations of the inputs from L4 were more consistent across PV cells from enucleated mice than in control animals. Together, our results show that after bilateral enucleation the AMPAR-mediated connections from L2/3 to L4 in PV interneurons are increased at P16/P17 and that the spatial pattern of connections from L4 and L5/6 was altered. Thus, similar to our results in Gad2 interneurons, the maturation and spatial refinement of AMPARmediated connections to PV interneurons in L4 of V1 requires retinal input, suggesting that peripheral input might be required for the maturation of different types of GABAergic interneurons.

NMDAR-only inputs are not altered in PV interneurons after enucleation. Binocular enuclea-
tion did not alter the NMDAR-only inputs to L4 Gad2 interneurons identified by our method. Thus, we investigated if NMDAR-only inputs received by PV interneurons were altered after enucleation. We then generated spatial connection probability maps (Fig. 7A) and laminar input plots (Fig. 7B) for NMDAR-only inputs to PV interneurons. Qualitative inspection showed that few NMDAR-only inputs were present but that the lami-  www.nature.com/scientificreports/ nar pattern of NMDAR-only inputs was highly variable. Quantifying the total number of NMDAR-only inputs showed no significant changes in the number of input locations in PV cells from control and enucleated animals (Fig. 7C, left; control median: 12, IQR: 8, enucleation median: 16, IQR: 7, P = 0.568, effect size: r = 0.11, Wilcoxon rank-sum test). Quantifying of the laminar sources of inputs showed that PV cells from enucleated and control animals received inputs from a similar number of locations from each layer (Fig. 7C, right; Table 7), but that the average proportion of NMDAR-only inputs from L2/3 was decreased ( Fig. 7D; Table 7). Analyzing the correlations of the input locations of NMDAR-only inputs revealed that the location of NMDAR-only inputs had similar variance in the control and enucleated animals ( Fig. 7E; Table 7). Calculating the overall fraction of NMDAR-only input revealed that the overall NMDAR-only input to AMPAR input ratio is comparable between control and enucleation group (Fig. 7F, left; control median: 0.2, IQR: 0.44, enucleation median: 0.18, IQR: 0.16, P = 0.462, effect size: r = 0.15, Wilcoxon rank-sum test), but that the NMDAR-only input to AMPAR input ratio was slightly decreased in L2/3 (Fig. 7F, right; Table 7).
Together, our data suggest that retinae are not required for the overall development of NMDAR-only inputs to L4 PV interneurons in V1 by P16/P17, but that enucleation might have a slight effect on inputs from L2/3.

The strength of AMPAR-mediated and NMDAR-only connections in PV interneurons is not altered after enucleation.
We next investigated if the strength of AMPAR-mediated inputs received by PV interneurons was affected by enucleation. Computing an average spatial map of AMPAR-mediated EPSC transferred charge revealed a pattern that was similar to the connection probability maps (Fig. 8A). The average EPSC transferred charge (Fig. 8B, left; control median: 0.47 pC, IQR: 0.19 pC, enucleation median: 0.55 pC, IQR: 0.17 pC, P = 0.086, effect size: Cohen's d = 0.71, two-sample t-test) of the evoked EPSCs was not changed. The average EPSC transferred charge in each layer was also similar between control and enucleated animals (Fig. 8B, right; Table 8). The average spatial map of the transferred charge of NMDAR-only EPSCs also showed no obvious differences between conditions (Fig. 8C). The average EPSC transferred charge (Fig. 8D, left; control median: 0.25 pC, IQR: 0.08 pC, enucleation median: 0.2 pC, IQR: 0.26 pC, P = 0.265, effect size: r = 0.22, Wilcoxon rank-sum test) was similar between PV cells from enucleated and control animals. When separated into different layers, the average EPSC transferred charge was also similar between the control and enucleated animals (Fig. 8D, right; Table 8). Thus, early removal of retinae has no effect on the strength of either AMPARmediated or NMDAR-only inputs to PV interneurons.

Discussion
Our results reveal that retinae are required for the normal development of AMPAR-mediated glutamatergic connections received by L4 GABAergic interneurons in the V1 during the pre-critical period of development. We show that in both Gad2-and PV-expressing interneurons the proportion of AMPAR-mediated inputs from L5/6 was reduced after enucleation, suggesting a similar dependence of glutamatergic circuit maturation on retinal input across different types of GABAergic interneurons. Thus, as a result of early retinal removal the translaminar organization of AMPAR-mediated circuit to L4 GABAergic neurons is shifted towards receiving more inputs from L2/3. However, Gad2 interneurons showed fewer inputs from L5/6 after enucleation while PV interneurons showed an overabundance of connections from L2/3. Moreover, these changes in the number of connections were accompanied by changes in the spatial pattern of connections. Thus, peripheral retinal input is required for the maturation of the translaminar organization of AMPAR-mediated circuits to L4 GABAergic Table 6. Statistics in Fig. 6. IQR interquartile range.  www.nature.com/scientificreports/ interneurons and the spatial patterning of these circuits. In contrast to the changes in AMPAR-mediated connections, NMDAR-only connections identified by our method were largely unaffected by enucleation. Our finding that the development of AMPAR-mediated connections to GABAergic Gad2 and PV interneurons in V1 is affected by peripheral retinal input is consistent with prior studies of excitatory inputs to GABAergic interneurons 23,51 . In rodents during the first two postnatal weeks, connections from cortical glutamatergic to GABAergic interneurons transition from initially being dominated by NMDAR-only synapses towards AMPARcontaining synapses [21][22][23] . Removal of cochlea changed the spatial distribution of AMPAR-mediated connections from L2/3 and L5/6 to Gad2 interneurons in the auditory cortex. Connections from L2/3 became more widespread while connections from L5/6 became more spatially restricted 23 . Blocking NMDAR-mediated thalamic activity in Reelin interneurons resulted in a loss of thalamic but increase in intracortical connectivity 51 , suggesting a peripheral activity-dependent reorganization of inputs. Our results here, showing that the loss of peripheral retinal input results in a laminar reorganization of AMPAR-mediated connections in V1, are consistent with these results. Thus, peripheral input is important for the normal development of the cortical GABAergic interneurons. However, due to the direct activation of the target cell by photolysis, our method cannot resolve connections within about 100 μm of the recorded neurons. It thus remains unknown if there are any changes to these very local connections after binocular enucleation.
Glutamatergic synapses often transition from a NMDAR-only state to an AMPAR-containing state during maturation 44,46,47 . This transition is associated with the closure of ocular dominance plasticity in the visual cortex 53,54 . We find that NMDAR-only connections are largely unaffected by peripheral manipulations, suggesting that NMDAR-only connections are regulated independently of AMPAR-mediated connections. However, our method can only detect NMDAR-only connections from locations distinct from the locations with AMPARmediated connections, so the NMDAR-only inputs we detected were underestimated. We cannot rule out the  www.nature.com/scientificreports/ possible changes of NMDAR-only inputs that are spatially located within the range of AMPAR-containing inputs-e.g., if presynaptic neurons in a particular location signal to the GABAergic interneuron under study via both AMPAR-containing and NMDAR-only synapses. Nevertheless, our results do suggest that the impact of enucleation is mainly on AMPAR-mediated connections.
Our results also showed that the excitatory neural networks innervating Gad2 and PV interneurons was differentially impacted by enucleation. We found that after enucleation L4 Gad2 interneurons received fewer AMPAR-mediated inputs from L5/6, while L4 PV interneurons received more AMPAR-mediated inputs from L2/3 and L4. The analysis of the spatial correlation of input locations revealed additional changes of the presynaptic locations in the presynaptic layers that did not show overt changes in input number in both Gad2 and PV interneurons. The locations of AMPAR-mediated inputs received by Gad2 interneurons became more heterogeneous, while for PV interneurons, the heterogeneity of the locations of AMPAR-mediated inputs decreased in L4 and increased in L5/6 after enucleation. The differences between Gad2 and PV interneurons likely reflect the heterogeneity of the Gad2 positive population 32,43 . GABAergic interneurons have diverse morphologies, electrical properties and synaptic connectivity 55,56 . Different GABAergic interneurons can vary in their developmental trajectory [48][49][50] , and respond differently to the same manipulation 51,57 . In contrast to the Gad2-positive population, the PV-Cre-labeled interneurons specifically encompass fast spiking interneurons 52 . Our results show that for both Gad2 and PV interneurons, binocular enucleation resulted in an increased proportion of AMPAR-mediated inputs from L2/3 and decreased proportion of AMPAR-mediated inputs from L5/6. These observations indicate that the development of cortical subgranular circuits and subgranular circuits might be differentially impacted by early removal of retinal input and thus might be differentially regulated. Subgranular layers encompass L6, which can be thalamorecipient [58][59][60][61][62][63][64][65] and also contain upper subplate neurons [66][67][68] . This suggests that pathways relaying peripheral activity engage these subgranular circuits and that subgranular circuits might play a key role during development.
Enucleation, widely used to study the influence of the sensory periphery on central visual circuits [69][70][71][72] , removes both early spontaneous and later visually driven retinal activity. In our experiments, binocular enucleation was done at P1/P2 and cortical glutamatergic inputs received by GABAergic interneurons were tested at P16/P17, when V1 activity patterns have matured 29 . This time period is before the onset of the ocular dominance plasticity in V1 40-42 but after eye opening. Before eye opening, the retina produces spontaneous activity [73][74][75][76] . At later stages and especially after eye opening retinal activity can be light-driven 29,77 . During this time period, retinal activity can affect neural activity in the developing visual cortex 29,78-81 and thus can contribute to the development of the visual cortex 5,82,83 . Thus, both spontaneous retinal waves and light-driven retinal activity could contribute to the development of AMPAR-mediated inputs received by L4 Gad2 and PV interneurons in V1. Further studies using selective manipulations of either spontaneous or light-evoked retinal activity are needed to evaluate the contributions of different sources of activity. Table 7. Statistics in Fig. 7. IQR interquartile range.  www.nature.com/scientificreports/ It is still unknown how much of the activity in V1 is changed at P16/17 after neonatal enucleation. In adult mice, after losing retinal input, neural activity in V1 rapidly decreases, recovering within 72 h due to homeostatic plasticity 70,84 . However, the activity in cortical inhibitory neurons does not recover within 72 h 85 , suggesting a compensatory decrease of cortical inhibition to decreased activity from retina. Similar homeostatic changes could happen in the neonatal enucleation model during early postnatal development. Peripheral manipulations  Table 8. www.nature.com/scientificreports/ can cause changes at multiple levels of processing, making direct inferences about the mechanisms underlying the effects we observe difficult. For example, correlations in LGN activity increase after optic nerve cuts 86 and such changes could at least partially compensate for retinal loss. Thus, changes seen on the cortical level could be due to changed average activity, changed correlations, or both in thalamocortical and possibly corticothalamic circuits. Nevertheless, our results show that the retina is required for the maturation of AMPAR-mediated connections onto GABergic interneurons. Peripheral input does not only influence GABAeric interneurons at early ages. Later manipulations of sensory activity can further change these connections. For example, monocular deprivation at P28/29 reduces the AMPAR-mediated inputs to PV interneurons in V1 87 . In particular, these studies showed that reduced visual activity led to decreased numbers of inputs and decreased amplitude of excitatory L4 and L5a input to L2/3 PV interneurons. Together with our data these results suggest that peripheral retinal input has a powerful influence on the laminar distribution of excitatory inputs onto inhibitory neurons over development, starting at the earliest postnatal ages.
In summary, our results show that early retinal input can shape the functional connectivity impinging on GABAergic interneurons. It is likely that early effects can alter the subsequent trajectory of cortical development.