Dynamic postnatal development of the cellular and circuit properties of striatal D1 and D2 spiny projection neurons

Key points Imbalances in the activity of the D1‐expressing direct pathway and D2‐expressing indirect pathway striatal projection neurons (SPNs) are thought to contribute to many basal ganglia disorders, including early‐onset neurodevelopmental disorders such as obsessive–compulsive disorder, attention deficit hyperactivity disorder and Tourette's syndrome. This study provides the first detailed quantitative investigation of development of D1 and D2 SPNs, including their cellular properties and connectivity within neural circuits, during the first postnatal weeks. This period is highly dynamic with many properties changing, but it is possible to make three main observations: many aspects of D1 and D2 SPNs progressively mature in parallel; there are notable exceptions when they diverge; and many of the defining properties of mature striatal SPNs and circuits are already established by the first and second postnatal weeks, suggesting guidance through intrinsic developmental programmes. These findings provide an experimental framework for future studies of striatal development in both health and disease. Abstract Many basal ganglia neurodevelopmental disorders are thought to result from imbalances in the activity of the D1‐expressing direct pathway and D2‐expressing indirect pathway striatal projection neurons (SPNs). Insight into these disorders is reliant on our understanding of normal D1 and D2 SPN development. Here we provide the first detailed study and quantification of the striatal cellular and circuit changes occurring for both D1 and D2 SPNs in the first postnatal weeks using in vitro whole‐cell patch‐clamp electrophysiology. Characterization of their intrinsic electrophysiological and morphological properties, the excitatory long‐range inputs coming from cortex and thalamus, as well their local gap junction and inhibitory synaptic connections reveals this period to be highly dynamic with numerous properties changing. However it is possible to make three main observations. Firstly, many aspects of SPNs mature in parallel, including intrinsic membrane properties, increases in dendritic arbours and spine densities, general synaptic inputs and expression of specific glutamate receptors. Secondly, there are notable exceptions, including a transient stronger thalamic innervation of D2 SPNs and stronger cortical NMDA receptor‐mediated inputs to D1 SPNs, both in the second postnatal week. Thirdly, many of the defining properties of mature D1 and D2 SPNs and striatal circuits are already established by the first and second postnatal weeks, including different electrophysiological properties as well as biased local inhibitory connections between SPNs, suggesting this is guided through intrinsic developmental programmes. Together these findings provide an experimental framework for future studies of D1 and D2 SPN development in health and disease.


Introduction
The striatum is the main input nucleus of the basal ganglia and consists of two populations of projection neurons with distinct long-range outputs, the D1-expressing direct pathway spiny projection neurons (SPNs) and the D2-expressing indirect pathway SPNs (Day et al. 2008;Gertler et al. 2008), which differentially regulate motor behaviour and cognitive function (Graybiel et al. 1994;Grillner et al. 2005;Yin & Knowlton, 2006;Kravitz et al. 2010;Tecuapetla et al. 2016). Adult D1 and D2 SPNs exhibit distinct electrical and morphological properties (Gertler et al. 2008) and form precise non-random local synaptic connections with each other (Taverna et al. 2008;Planert et al. 2010;Cepeda et al. 2013). Imbalance in the activity of the two pathways is thought to contribute to the cognitive and motor symptoms seen in late onset neurodegenerative disorders such as Parkinson's disease (Taverna et al. 2008) and Huntington's disease (Cepeda et al. 2013), but also those seen in early onset neurodevelopmental disorders such as Tourette's syndrome (McNaught & Mink, 2011;Albin, 2018), obsessive-compulsive disorder (Graybiel & Rauch, 2000;Langen et al. 2011), attention deficit hyperactivity disorder (Del Campo et al. 2011) and autism spectrum disorders (Shepherd, 2013). The cellular and neural circuit changes that underpin these neurodevelopmental disorders are major research areas. Although key papers have started to shed light on early postnatal striatal development Dehorter et al. 2011;Kozorovitskiy et al. 2012;Peixoto et al. 2016), often SPNs have been grouped together as one population and therefore many aspects of D1 and D2 SPN postnatal development remain unknown.
A combination of whole-cell patch-clamp electrophysiology and anatomical analysis in mouse brain slices allows for the investigation of the cellular and circuit properties of striatal D1 and D2 SPNs from the earliest postnatal periods into maturity. These include postnatal day (P)3-6, the period when most striatal SPNs have been born but excitatory synaptic input to the striatum is thought to be minimal and mouse pups produce little movement; P9-12, when excitatory synaptic inputs to the striatum are thought to have undergone a period of rapid maturation and motor competence of the pups has increased; P21-28, when the striatal neurons and the circuit are approaching maturity and mice readily traverse the environment; and finally P35+, when the brain is thought to have reached maturity coinciding with the sexual maturity of mice (Finlay & Darlington, 1995;Tepper et al. 1998;Khazipov et al. 2004;Dehorter et al. 2011;Kozorovitskiy et al. 2012;Peixoto et al. 2016). Overall, we found that the early postnatal development of striatal D1 and D2 SPNs is highly dynamic with many intrinsic and circuit properties changing. We found that young D1 SPNs are electrophysiologically more mature than D2 SPNs and that intrinsic electrophysiological differences between adult D1 and D2 SPNs are already apparent in the second postnatal week. Both D1 and D2 SPNs exhibit similar increases in dendritic arbour and spine density and equally sample excitatory cortical and thalamic inputs in the first postnatal week. Subsequent maturation of excitatory synapses occurs mostly in parallel and is relatively rapid for thalamic synapses and more prolonged for cortical synapses. The notable exception is a transient strong input to D2 SPNs from thalamus and a stronger NMDA receptor-mediated input to D1 SPNs, both in the second postnatal week. All excitatory inputs in the second postnatal week are further characterized by their long durations and decay times and pharmacological study suggests this is mediated through expression of specific combinations of glutamate receptors. Inhibitory synapses onto SPNs are initially sparser and exhibit a more prolonged maturation, as reflected by a progressive increase in miniature inhibitory postsynaptic current (mIPSC) frequency. Indeed, simultaneous quadruple patch-clamp recordings and the study of local connections between developing SPNs reveals that in the first postnatal week SPNs mainly form gap junctions with each other which only in later postnatal weeks are increasingly replaced by inhibitory synaptic connections. Interestingly, these early inhibitory synaptic connections are precise and non-random and relative biases in synaptic connectivity found in adulthood are already apparent in the second postnatal week, including highly interconnected D2 SPNs. Together, these results suggest that striatal D1 and D2 SPN postnatal development is both highly dynamic and organized with many of the cellular and circuit properties established soon after birth suggesting a role for intrinsic developmental programmes in guiding their early development.

Ethical approval
The present study conforms to the ethical principles and regulations of The Journal of Physiology and with The Journal's animal ethics checklist as described by Grundy (2015). All animal work performed at the University of Oxford (UK) was licensed by the Home Office under the Animals (Scientific Procedures) Act 1986 and was approved by the University of Oxford Ethical Review Committee. All efforts were taken to minimize animal numbers.

Animals
All experiments were carried out on C57/BL6 wild-type and heterozygous D1-GFP or D2-GFP mice of both sexes with ad libitum access to food and water. The D1-GFP or D2-GFP bacterial artificial chromosome (BAC) transgenic mice report subtypes of the dopamine receptor, either D1 or D2, by the presence of green fluorescent protein (GFP) (Mutant Mouse Regional Resource Centres (MMRRC), USA). Details of the mice and the methods of BAC mice production have been published (Gong et al. 2003) and can be found on the GENSAT website (GENSAT (2009) The Gene Expression Nervous System Atlas (GENSAT) Project. In: NINDS, Contracts N01NS02331 and HHSN271200723701C, The Rockefeller University (New York, USA), http://www.gensat.org/index.html). In brief, the genotype of the mice has been modified to contain multiple copies of a modified BAC in which the enhanced GFP (EGFP) reporter gene is inserted immediately upstream of the coding sequence of the D1 or D2 gene. These BAC transgenic mice arrived originally on a Swiss Webster background but were backcrossed to a C57/BL6 background over 20+ generations prior to use and kept as a heterozygous mouse line to avoid published issues using these transgenic lines (Bagetta et al. 2011;Kramer et al. 2011;Chan et al. 2012;Nelson et al. 2012). Experiments were designed to use litter mates for the various age ranges within single experiments so as to control for effects of litter sizes and maternal care factors that could affect levels of neuronal and circuit maturity. All mice were bred, individually ventilated cage (IVC) housed in a temperature controlled animal facility (normal 12:12 h light-dark cycles) and used in accordance with the UK Animals (Scientific Procedures) Act (1986).

Stimulation and recording protocols
Hyperpolarizing and depolarizing current steps were used to assess the intrinsic properties of the recorded SPNs J Physiol 597.21 including input resistance, spike threshold (using small incremental current steps) and membrane time constant, as well as the properties of action potentials (amplitude, frequency and duration). Properties were assessed immediately on break-in. Currents step ranges were for P3-6: −50 to +50 pA; for P9-12: −100 to +100 pA; and for P21-28 and P35+: −500 to +500 pA. These ranges of currents were chosen to allow sufficient depolarization of SPNs taking into consideration changes in input resistance and observations of depolarization block and action potential failure in SPNs. In a subset of neurons the membrane capacitance was assessed from the area under the capacitive transient as a result of repeated voltage steps ('seal test' , 5 mV, at 20 Hz) in voltage-clamp mode. Both miniature excitatory postsynaptic currents (mEPSCs) and mIPSCs were recorded from individual SPNs, held at respectively −75 and 0 mV in aCSF containing 1 µM TTX, in 5 min sweeps. Neurons were kept for 5+ minutes in whole-cell configuration mode before miniature recordings started to facilitate ionic and QX-314 diffusion of the intracellular solution. Activation of excitatory cortical and thalamic afferents was performed via a bipolar stimulating electrode (FHC Inc., Bowdoin, ME, USA) placed in respectively the external or internal capsule, and in the presence of blockers of inhibitory GABAergic transmission including the GABA A receptor antagonist SR95531 (1 µM) and the GABA B receptor antagonist CGP52432 (2 µM). Afferents were activated every 5 s with up to 20 repetitions and excitatory postsynaptic currents (EPSCs) and excitatory postsynaptic potentials (EPSPs) were recorded from the patched SPNs. Evoked EPSCs were recorded in whole-cell voltage-clamp mode at a holding potential near −75 mV and evoked EPSPs in whole-cell current-clamp mode at resting membrane potential. Trains of stimulations consisted of 10 pulses given at 20 Hz and trains were repeated every 30 s up to 5 times. Combined AMPA/kainate and NMDA receptor-mediated currents were recorded from SPNs held at +50 mV. AMPA/kainate receptor-mediated currents were recorded after a 5-10 min wash-in of the NMDA receptor antagonist D-(−)-2-amino-5-phosphonopentanoic acid (D-AP5; 50 µM). The contribution of different glutamate receptor subtypes to striatal evoked EPSPs was investigated using superfusion of the NR2C/D subunit-selective NMDA receptor antagonist 2S * ,3R * -1-(Phenanthren-2-carbonyl)piperazine-2,3-dicarboxylic acid (PPDA) (200 nM), the NMDA receptor antagonist D-AP5 (50 µM), the AMPA/kainate receptor antagonist 2,3-dihydroxy-6-nitro-7-sulfamoylbenzo[f]quinoxaline-2,3-dione (NBQX; 20 µM) and the kainate receptor antagonist UBP-310 (5 µM). All drugs were obtained from Tocris Biosciences (Bristol, UK). Local gap junctions between SPNs were examined by delivering hyperpolarizing current injections (200 ms, P3-6: −20 pA; P9-12: −100 pA; and P21-28: −200 pA) to each patched SPN sequentially, whilst simultaneously monitoring the membrane voltage of the other SPNs. Local inhibitory synaptic connectivity between SPNs was examined by delivering brief (ß60 ms) suprathreshold current injections (P3-6: +50 pA; P9-12: +150 pA; and P21-28: +400 pA) or brief trains of current injections (6 pulses, 30 ms, P3-6: +80 pA; P9-12: +200 pA; and P21-28: +500 pA at 20 Hz) to each patched SPN sequentially, whilst simultaneously monitoring the membrane voltage of the other SPNs. Protocols were repeated 20-30 times for the detection of gap junctions and synaptic connections.

Analysis of recordings
Data were analysed offline using custom written programmes in Igor Pro (Wavemetrics, Lake Oswego, OR, USA, RRID:SCR_000325). The input resistance was calculated from the observed membrane potential change after hyperpolarizing the membrane potential with a set current injection. The membrane time constant was taken as the time it takes for a change in potential to reach 63% of its final value. The spike threshold was the membrane voltage at which the SPN generated an action potential. The action potential amplitude was taken from the peak amplitude of the individual action potentials relative to the average steady-state membrane depolarization during positive current injection. Action potential duration was taken as the duration between the upward and downward stroke of the action potential at 25% of the peak amplitude. mEPSCs and mIPSCs were detected as downward and upward deflections of more than 2 standard deviations (SD) above baseline (baseline consisted of the average holding current across the entire recording) and more than 10 ms duration in 5 min duration traces which were lowpass filtered at 50 Hz. Miniature events were not corrected for developmental changes in membrane capacitance. Evoked EPSCs, EPSPs and inhibitory postsynaptic potentials (IPSPs) were defined as upward or downward deflections of more than 2 SD on average synaptic responses generated after filtering and averaging original traces (0.1 Hz high-pass filter and 500 Hz low-pass filter) and used for analysis of synaptic properties. Synaptic properties include measurements of peak amplitude, duration (measured from the start of the upward/downward stroke of the event until its return to the pre-event baseline), rise time (time between 20% and 80% of the peak amplitude) and decay time (measured as the time from peak amplitude until the event returned to 50% of peak amplitude). Synaptic delays were calculated from the time of stimulation to the start of the upward stroke of the synaptic response. The short-term plastic properties of cortical and thalamic excitatory synapses and inhibitory synapses between SPNs were analysed by taking the amplitude of each EPSP/IPSP during train stimulation and dividing this by the amplitude of the first response. The NMDA/AMPA ratio was calculated from recordings of the combined AMPA/ kainate and NMDA receptor-mediated current as well as the pharmacologically isolated AMPA/kainate receptor-mediated current. The average AMPA/kainate receptor-mediated current trace was subtracted from the combined AMPA/kainate and NMDA receptor-mediated current trace to obtain the NMDA receptormediated current. Peak amplitude NMDA receptormediated current was divided by peak amplitude AMPA/kainate receptor-mediated current to obtain the NMDA/AMPA ratio. The presence of gap junctions was assessed by averaging the 20-30 sweeps consisting of hyperpolarizing current injections and observing a significant downward deflection of more than 2 SD from baseline. The coupling coefficient (CC) was obtained by dividing the amplitude of the low-frequency voltage change in the receiver SPN to that in the driver SPN. The junctional conductance (G j ) was estimated from R input and CC (Venance et al. 2004 where R input1 and R input2 are the R input values of the injected and receiving SPNs, respectively, and CC 1-2 the CC between the injected and receiving SPNs.

Histological analyses and cell classification
Following whole-cell patch-clamp recordings, the brain slices were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer (PB; pH 7.4). Biocytin-filled neurons were visualized by incubating sections in 1:10,000 streptavidin AlexaFluor405-conjugated antibodies (Thermo Fisher Scientific, Waltham, MA, USA, cat. no. S32351). Visualized neurons were labelled for chicken ovalbumin upstream promoter transcription-factor interacting protein-2 (CTIP2; 1:1000, rat, Abcam (Cambridge, UK), cat. no. ab14865, RRID:AB_2064130) and pre-proenkephalin (PPE; 1:1000, rabbit, LifeSpan Biosciences (Seattle, WA, USA), cat. no. LS-C23084, RRID:AB_902714) in PBS containing 0.3% Triton X-100 (PBS-Tx) overnight at 4°C followed by incubation with goat anti-rat AlexaFluor647 (1:500; Thermo Fisher Scientific, cat. no. A-21247, RRID:AB_141778) and goat anti-rabbit AlexaFluor555 (1:500; Thermo Fisher Scientific, cat. no. A-21429, RRID:AB_2535850) secondary antibodies in 0.3% PBS-Tx for 2 h at room temperature for D1 or D2 SPN classification. CTIP2 is expressed by SPNs and not interneurons (Arlotta et al. 2008) and PPE reliably labels indirect pathway D2 SPNs (Lee et al. 1997;Sharott et al. 2017). PPE antibody staining was facilitated through antigen retrieval by heating sections at 80°C in 10 mM sodium citrate (pH 6.0) for approximately 30-60 min prior to incubation with PPE primary antibody. After classification of SPNs the slices were washed 3 times in PBS and processed for 3,3 -diaminobenzidine (DAB) immunohistochemistry using standard procedures. Fluorescence images were captured with a LSM 710 confocal microscope using ZEN software (Zeiss, Cambridge, UK; RRID:SCR_013672). DAB-immunoreactive neurons were visualized on a brightfield microscope and were reconstructed and analysed using Neurolucida and Neuroexplorer software (MBF Bioscience, Delft, The Netherlands; RRID: SCR_001775). Only labelled neurons that exhibited a full dendritic arbour were included for analysis, e.g. cells with clear truncations were not included in the dataset. Scholl analysis and polarity analysis were performed using standard procedures. In brief, both Scholl and polarity plots were generated for individual SPNs by calculating the total dendritic length located within 10°segments with increasing distance from the soma. The dendritic lengths were subsequently normalized for an individual SPN and averaging the normalized plots of individual neurons generated final plots.

Statistics
All data are presented as means ± SEM; n refers to the number of neurons tested. The following numbers of animals were used for the datasets as reported in Fig . Gap junction and synaptic connection incidence were compared using Fisher's exact test. Continuous data were assessed for normality and appropriate parametric (ANOVA, Student's paired t test and unpaired t test) or non-parametric (Mann-Whitney U) statistical tests were applied ( * P < 0.05, * * P < 0.01, * * * P < 0.001).
Secondly, both D1 and D2 SPNs exhibited a progressive maturation of their intrinsic membrane properties, including the emergence of a pronounced inward rectifying current at later developmental stages (Fig. 1D), a more hyperpolarized membrane potential (Fig. 1E) and a lowering of input resistance (Fig. 1F). The higher action potential frequency seen in D2 SPNs might well result from their consistently more depolarized membrane potential (P9-12: D1: −71.0 ± 0.9 mV and D2: In conclusion, we found that most D1 and D2 SPNs are able to generate action potentials shortly after birth and many of their electrophysiological properties develop in parallel. However, maturational differences could be seen early on, including narrower and larger action potentials in the D1 SPNs. Furthermore, significant differences in the intrinsic membrane properties were already observed in the first and second postnatal weeks, which persisted into adulthood, including a greater excitability and action potential frequency of D2 SPNs.
In conclusion, we found that the general morphology of the D1 and D2 SPNs develops in parallel with similar increases in their dendritic arbour and spine density.

Maturation of excitatory and inhibitory synaptic inputs onto D1 and D2 SPNs
Our results so far suggest that both D1 and D2 SPNs can already generate action potentials during the first postnatal week, and that their dendritic arbour and spine density develop mostly in parallel, allowing them to sample excitatory and inhibitory synaptic inputs from nearby axons. We next asked when synaptic inputs on SPNs are functional by performing whole-cell voltage-clamp recordings of SPNs in the presence of TTX (1 µM) at the four age ranges. This allowed for recordings of both spontaneous mEPSCs by holding the SPN membrane voltage at −70 mV (Fig. 3A) and spontaneous mIPSCs by holding the SPN membrane voltage at 0 mV (Fig. 3D). We confirmed that spontaneous miniature events could be blocked using respectively the AMPA/kainate receptor antagonist NBQX (10 µM) and the GABA A receptor antagonist SR95531 (200 nM) ( Fig. 3A and D). Our first observation was that excitatory mEPSCs could be detected as early as P3-6 in both D1 and D2 SPNs ( Fig. 3A and B), which increased slightly in frequency but was already close to that seen in adulthood (ß1 Hz) (P3-6: 0.79 ± 0.10 Hz; P9-12: 0.88 ± 0.07 Hz; P21-28: 1.05 ± 0.10 Hz; and P35+: 0.89 ± 0.10 Hz; P3-6 vs. P35+ P = 0.518, Mann-Whitney U test, n = 15, 22, 17 and 14; Fig. 3B). In contrast, the mEPSC amplitude exhibited a significant increase for both D1 and D2 SPNs from P3-6 to P9-12 (D1: P3-6: 3.06 ± 0.50 pA to P9-12: 7.13 ± 0.57 pA; and D2: P3-6: 2.52 ± 0.40 pA to P9-12: 6.67 ± 0.47 pA; P = 0.001 and P = 0.001, Mann-Whitney U test, both n = 6 and n = 7; Fig. 3C) after which it remained constant. Importantly, no significant differences were found in either the mEPSC J Physiol 597.21 frequency or the mEPSC amplitude between the D1 and D2 SPNs at any of the age ranges investigated (all P > 0.05). These results suggest that excitatory synaptic inputs on SPNs are present and functional soon after birth and develop in parallel and similarly innervate both D1 and D2 SPNs with postsynaptic changes occurring between P3-6 and P9-12.
These results suggest that functional excitatory and inhibitory synaptic inputs are present during the first postnatal week and are sampled by both D1 and D2 SPNs. Moreover, they suggest that between the first and second postnatal weeks, substantial postsynaptic changes occur as reflected in the greater mEPSC and mIPSC amplitudes. Lastly, whereas the mEPSC frequency stayed relatively constant, a progressive and steady increase in mIPSC frequency was seen implying a prolonged maturation of inhibitory inputs.

Figure 2. Development of dendritic arbours and spines of D1 and D2 SPNs
A, example reconstructions of previously recorded SPNs processed for DAB immunohistochemistry. SPNs are grouped according to age (left to right, P3-6, P9-12, P21-28 and P35+) and whether they are D1 (orange, top) or D2 (blue, bottom) SPNs. The examples shown are all reconstructed and analysed neurons from coronal sections and are aligned such that top is dorsal, bottom is ventral, left is lateral and right is medial. B, D1 (orange) and D2 (blue) SPNs exhibit a significant and similar increase in their dendritic length as they mature C, Scholl analysis of dendritic complexity of D1 and D2 SPNs reveals a similar elaboration of distal dendritic segments as they mature. D, polarity analysis of dendrites of D1 and D2 SPNs reveals a mostly uniform and radial distribution of their dendrites. Note the bias to extend dendrites from lateral-ventral aspects to medial-dorsal aspects. E, spine density of D1 and D2 SPNs in different age ranges. Note the similar increase in spine density in both D1 and D2 SPNs as they mature.
To investigate whether the observed increase in amplitude of the evoked excitatory response could be the result of changes in postsynaptic glutamate receptors, we analysed the contribution of both NMDA and AMPA/kainate glutamate receptors to the cortically evoked excitatory responses at the different age ranges. Evoked excitatory events were recorded from SPNs in whole-cell voltage-clamp mode at a holding potential of +50 mV and consisted of combined NMDA and AMPA/kainate receptor-mediated currents (Fig. 4D). After baseline recording, the NMDA receptor antagonist D-AP5 (50 µM) was superfused thereby isolating the AMPA/kainate receptor-mediated current. Analysis of the ratio of peak amplitude NMDA and AMPA/kainate receptor-mediated currents revealed a decline in the NMDA/AMPA ratio across early development, especially evident from P9-12 to P21-28 (P3-6: 2.48 ± 0.49; P9-12: 2.15 ± 0.30; P21-28: 1.00 ± 0.39; and P35+: 1.43 ± 0.21; P3-6 vs. P21-28: P = 0.004, Mann-Whitney U test, n = 16, 20, 12 and 11; Fig. 4D) and similarly for both the D1 and D2 SPNs ( Fig. 4D and Table 2).
Such changes in postsynaptic glutamate receptor types are predicted to change the EPSP kinetics (Seeburg, 1993). Indeed, we found that both the EPSP duration and the EPSPs between the D1 (orange) and D2 (blue) SPNs. Also, note that ß70% of SPNs exhibited a response at P3-6 and responses could always be observed at later ages. C, bar plot of the maximum evoked EPSP amplitude across the age ranges which remains relatively constant at ß3 to 4 mV (top). Bar plot of EPSC amplitude shows an increase in the cortically evoked excitatory current (bottom), especially evident from P9-12 to P21-28. D, bar plots of the NMDA/AMPA ratio across the age ranges. Note the decrease in the ratio as the neurons mature (top), which occurs in parallel for both D1 and D2 SPNs (bottom). E, bar plots of the NMDA receptor-mediated current (top) and AMPA/kainate receptor-mediated current (bottom) across the age ranges. Note the significant increase in the AMPA/kainate receptor-mediated current whereas the NMDA receptor-mediated current stayed constant. F, bar plots of the EPSP duration and decay time. Note the transient and significant increase in the EPSP duration and decay time at P9-12. G, graphs of the EPSP amplitude across 10 stimulations at 20 Hz showing that corticostriatal synapses at D1 and D2 SPNs predominantly exhibit short-term depression at all age ranges. D1 SPNs: orange squares; D2 SPNs: blue squares; and unclassified SPNs: grey squares.    4F and Table 2). These developmental changes were also reflected in the kinetics of the EPSCs (Table 2; EPSC duration: P3-6 vs. P9-12: P = 0.012; and P9-12 vs. P21-28: P = 0.042; and EPSC decay time: P3-6 vs. P9-12: P = 0.0089; and P9-12 vs. P21-28: P = 0.070). Neither the EPSP nor the EPSC kinetics differed between the D1 and D2 SPNs (Table 2; all P > 0.05). Lastly, we investigated whether there were also presynaptic developmental changes that occurred at corticostriatal synapses, which could affect the short-term plastic properties of the cortical synapses onto SPNs. Using trains of electrical stimulation (10 pulses at 20 Hz) we found that corticostriatal synapses were consistently depressing at all developmental ages (Fig. 4G).
Combined, these results suggest that the excitatory cortical synapses onto D1 and D2 SPNs are functional in the first postnatal week and mostly develop in parallel and become stronger across the postnatal weeks, mainly from P9-12 onwards through an increase in AMPA receptor-mediated transmission. The notable exception is a larger NMDA receptor-mediated current in D1 SPNs in the second postnatal week. Lastly, we found that corticostriatal excitatory responses exhibit both a long duration and decay time in the second postnatal week.

Maturation of long-range cortical thalamic inputs on striatal D1 and D2 SPNs
The second major excitatory input to the striatal SPNs comes from the thalamus (Doig et al. 2010;Ellender et al. 2013;Smith et al. 2014), whose inputs are thought to arrive comparatively earlier in development (Nakamura et al. 2005). To investigate the development of the excitatory inputs coming from thalamus, we performed whole-cell patch-clamp recordings of D1 and D2 SPNs in the dorsal striatum and activated excitatory afferents from the thalamus by giving single and trains of stimulation via a tungsten bipolar electrode placed in the internal capsule (Fig. 5A). These experiments were performed in modified horizontal sections, to retain as much of the thalamostriatal projections as possible (Ding et al. 2008;Smeal et al. 2008), and in the presence of GABA receptor antagonists to avoid erroneous activation of inhibitory inputs. Similar to our observations for cortical inputs, not all D1 and D2 SPNs at P3-6 received thalamic inputs (D1: 64% and D2: 77%, n = 14 and 13), whereas those at P9-12 and older all did (Fig. 5B). For all SPNs receiving thalamic input (mean synaptic delay of ß4 ms; P3-6: 4.31 ± 0.35 ms; P9-12: 3.78 ± 0.25 ms; P21-28: 3.69 ± 0.30 ms; and P35+: 3.89 ± 0.54 ms) we found that across a wide range of stimulation strengths (range 20-220 µA) both D1 and D2 SPNs mostly received comparable amplitude EPSPs, with the notable exception of D2 SPNs, which transiently receive a larger thalamic excitatory input at P9-12 (F(1, 79) = 6.726, P = 0.011; Fig. 5B). This was also reflected in larger amplitude EPSCs as recorded from D2 SPN at P9-12 (D1: 20.32 ± 7.72 pA and D2: 49.17 ± 6.67 pA, P = 0.013, Mann-Whitney U test, n = 14 and 9; Table 3). In this case and others, the distance between the stimulation electrode and recording electrode was kept constant for both D1 and D2 SPNs J Physiol 597.21
Interestingly, we observed a drop in both NMDA and AMPA/kainate receptor-mediated currents in the P35+ range but this did not reach significance.
Lastly, we investigated whether presynaptic changes could be observed at thalamostriatal synapses that could affect the short-term plastic properties of the thalamic synapses onto SPNs. Using trains of electrical stimulation (10 pulses at 20 Hz) we found that thalamic synapses were consistently depressing at all developmental ages but exhibited more pronounced depression on D2 SPNs at P9-12 (P = 0.015) and on D1 SPNs at P21-28 (P = 0.0001, Fig. 5G).
Combined, these results suggest that the excitatory thalamic synapses onto D1 and D2 SPNs are also functional in the first postnatal week and mostly develop in parallel with the second postnatal week also characterized by long duration EPSPs. The notable exception is a transient stronger thalamic input to D2 SPNs in the second postnatal week. Furthermore we found that, in contrast to the cortical synapses, the thalamostriatal synapses exhibit a rapid change in the NMDA/AMPA ratio as a result of larger increases in AMPA/kainate receptor-mediated currents from the first to second postnatal week.
In conclusion, we found that glutamate receptor expression at excitatory synapses onto both D1 and D2 SPNs changes across development and can differentially affect the EPSP kinetics. The pharmacological experiments would suggest that in the first postnatal week excitatory synapses contain many NR2C/D subunit-containing NMDA receptors, followed in the second postnatal week by a transient high expression of the NR2A/B subunit-containing NMDA receptors. Overall, the expression of AMPA-and kainate receptors seems to progressively increase across development.

Maturation of local inhibitory synaptic connections between striatal SPNs
Our data so far suggests that most of D1 and D2 SPNs can receive excitatory inputs from both the cortex and the thalamus, and are able to generate action potentials allowing them to signal to downstream basal ganglia nuclei during the first postnatal week. However, the probability and timing of these action potentials is under control of inhibition provided by both lateral inhibitory connections between SPNs and inputs from striatal interneurons (Tepper & Plenz, 2006;Ponzi & Wickens, 2010). The analysis of mIPSC frequency would suggest an extended and progressive increase in the number of inhibitory inputs across early postnatal development but whether these arise from neighbouring SPNs or interneurons is unknown. To investigate when SPNs form inhibitory synaptic connections with each other and how these connections change across development, we performed quadruple whole-cell current-clamp recordings of SPNs in the first four postnatal weeks (Fig. 7A) including post hoc immunocytochemistry (Fig. 7B) and histochemistry (Fig. 7C) to classify recorded neurons as putative D1 or D2 SPNs (see Methods). As immature SPNs have been shown to form gap junctions with each other (Venance et al. 2004), with which they can also regulate each other's activity, both hyperpolarizing and depolarizing current steps were used to investigate both gap junctions (Fig. 7D) and synaptic connections (Fig. 7E) between SPNs.

Figure 7. Gradual replacement of symmetrical gap junctions with precise inhibitory synaptic connections between SPNs
A, Dodt-contrast image of recording configuration consisting of four simultaneously patched SPNs. B, post hoc immunocytochemistry of recorded neurons using antibodies against streptavidin, PPE and CTIP2 allowed for classification of neurons as D1 or D2 SPNs. Note that SPN no. 1 is PPE negative and CTIP2 positive and therefore a D1 SPN, whereas SPN no. 2 is positive for PPE (indicated by asterisk) and therefore a D2 SPN. C, subsequently the slices were processed for DAB immunohistochemistry to label SPNs (left) and reveal dendritic structures allowing for reconstruction of SPNs (right). D, hyperpolarizing current steps revealed the presence of potential gap junctions connecting recorded SPNs. Note the presence of bidirectional gap junctions between D1 SPN no. 1 (orange) and D2 SPN no. 3 (blue). E, suprathreshold current injections elicited action potentials in recorded SPNs and revealed potential synaptic connections to other simultaneously recorded SPNs. Note the presence of a unidirectional synaptic connection from D2 SPN no. 2 to D2 SPN no. 3. F, diagram of experimental set-up to test for potential gap junctions between SPNs (left). Bar plots showing a significant decrease in the incidence of detected gap junctions as the SPNs mature (right). G, bar plots of the incidence of gap junctions between D1 and D2 SPNs across the age ranges. Note the relatively uniform incidence of gap junctions in all SPN groups at P3-6 followed by a progressive reduction and absence of detected gap junctions at P21-28. H, diagram of experimental set-up to test for synaptic connections between SPNs (left). Bar plots showing a progressive and significant increase in the incidence of detected synaptic connections as the SPNs mature (right). I, bar plots of incidences of synaptic connections between D1 and D2 SPNs across the age ranges. Note the earliest appearance of synaptic connections at P3-6 from D1 SPNs only. By P9-12 synaptic connections from both D1 and D2 SPNs can be observed and relative biases in synaptic connectivity, i.e. high incidence of connectivity between D2 SPN, are already apparent and are maintained. strongest synaptic connections (P9-12: 0.97 ± 0.32 mV and P21-28: 0.72 ± 0.28 mV; Table 5) (Planert et al. 2010).
Lastly, no significant differences in other properties of the synaptic connections were observed between the different SPN types or across the age ranges (Table 5).
Together, these results demonstrate that as the striatal circuit matures, symmetric gap junctions between both D1 and D2 SPNs are gradually replaced with precise unidirectional local inhibitory synaptic connections. Moreover, these inhibitory synaptic connections exhibit biases, such as the high incidence of connections between D2 SPNS, and are already established by the second postnatal week.

Discussion
In this paper we describe the developmental trajectory of identified D1 and D2 SPNs during the first postnatal weeks. We found that the striatal cellular and circuit properties are highly dynamic during this period but several general observations can be made. Firstly, young D1 SPNs are electrically more mature and intrinsic differences in the electrical properties of D1 and D2 SPNs are apparent by the second postnatal week and maintained into adulthood. Secondly, both D1 and D2 SPNs initially exhibit small radially oriented dendrites, which further develop in parallel including increases in length, complexity and spine density. Thirdly, we found that early excitatory synapses onto D1 and D2 SPNs are functional and indeed most SPNs receive long-range excitatory synaptic inputs from both cortex and thalamus in the first postnatal week. Both inputs progressively strengthen through dynamic changes in postsynaptic glutamate receptor expression, which occurs relatively rapidly for thalamic synapses. Furthermore, we found that excitatory synapses in the second postnatal week exhibit several unique features including a transient strong thalamic drive to D2 SPNs, a stronger NMDA receptor-mediated cortical input to D1 SPNs, as well as long duration EPSPs. Fourthly, although we found that inhibitory synapses onto D1 and D2 SPNs are functional in the first postnatal week, the development of inhibitory synaptic connections is overall more protracted. Indeed, initially SPNs communicate locally through gap junctions, which are progressively replaced by precise inhibitory synaptic connections in the second and later postnatal weeks. Interestingly, clear biases in inhibitory connections between D1 and D2 SPNs are already apparent in the second postnatal week and are maintained into adulthood. Overall, these findings suggest that early postnatal development of D1 and D2 SPNs follows a dynamic but organized trajectory with many of the cellular and circuit properties established soon after birth.

Intrinsic cellular properties of D1 and D2 SPNs
We found a progressive development of both the intrinsic electrophysiological and the morphological properties of the D1 and D2 SPNs. Both SPNs are able to generate small 'immature' action potentials in the first postnatal days and both undergo a progressive decrease in their input resistance and a hyperpolarization of their resting membrane potential (Lieberman et al. 2018), concurrent with an ability to generate large 'mature' action potentials at higher firing frequencies (Peixoto et al. 2016). However, some differences were observed. Initially, the size and duration of the action potentials were more mature for the D1 SPNs, possibly as a result of their J Physiol 597.21 suggested earlier birthdate (Marchand & Lajoie, 1986;van der Kooy & Fishell, 1987;Kelly et al. 2018). Secondly, many of the differences in the electrical properties of SPNs, such as the comparatively more depolarized resting membrane potential, higher input resistance and higher firing frequencies of D2 SPNs, are already apparent in the second postnatal week and are maintained into adulthood (Gertler et al. 2008;Peixoto et al. 2016;Lieberman et al. 2018). Morphologically we found that SPNs exhibit a radial dendritic morphology from birth, which undergoes a substantial elaboration concomitant with an increase in dendritic spine density occurring in parallel for the D1 and D2 SPNs. It was not possible to distinguish between D1 and D2 SPNs at any of the age ranges suggesting that previously described differences in morphology (e.g. the increased dendritic arbourization of D1 SPNs (Gertler et al. 2008;Benthall et al. 2018) or spine density (Gertler et al. 2008;Kozorovitskiy et al. 2012) might be due to specifics of age, mouse line or methodology (Bagetta et al. 2011;Kramer et al. 2011;Chan et al. 2012;Nelson et al. 2012). Overall we found a progressive increase in dendritic spine density consistent with previous studies Tepper et al. 1998), although total spine density is lower compared to two-photon imaging and serial section EM studies (Ingham et al. 1998;Day et al. 2006;Kozorovitskiy et al. 2012) likely due to our use of DAB immunohistochemistry and occlusion of spines on the top and bottom of dendrites by the DAB reaction product (Ingham et al. 1998;Day et al. 2006).

Functional excitatory synaptic inputs onto D1 and D2 SPNs
We found that already soon after birth excitatory inputs onto SPNs are functional and are able to depolarize both D1 and D2 SPNs. Whereas the frequency of mEPSCs in the first postnatal week is close to that observed in adulthood, a large increase in mEPSC amplitude is seen for both D1 and D2 SPNs from the first to second postnatal week, consistent with previous observations (Dehorter et al. 2009;Peixoto et al. 2016), and suggestive of postsynaptic changes. Indeed, in recordings of electrically evoked corticostriatal and thalamostriatal responses we found that the majority of D1 and D2 SPNs have functional synapses in the first postnatal week and both exhibit similar increases in amplitude (Day et al. 2006). This increase in response amplitudes correlates with larger AMPA/kainate receptor-mediated currents and concomitant decreases in the NMDA/AMPA ratio seen in both D1 and D2 SPNs (Colwell et al. 1998;Hurst et al. 2001;Peixoto et al. 2016). Interestingly, this occurs rapidly at thalamostriatal synapses from the first to second postnatal week whereas the corticostriatal synapses exhibit a more gradual maturation extending into later postnatal weeks. Overall we found that the second postnatal week exhibits many interesting features in that thalamic inputs to D2 SPNs are transiently larger in amplitude as well as D1 SPNs exhibiting larger NMDA receptor-mediated inputs from cortex, which could result from transient changes in receptor expression or differential maturation of inputs. Furthermore, the EPSP kinetics during this period are characterized by their long durations and decay times. Pharmacological study of glutamate receptor expression at these excitatory synapses would suggest that their expression is highly dynamic in these early postnatal weeks with a progressive decrease in NR2C/D subunit-containing NMDA receptors, consistent with previous observations (Monyer et al. 1994;Dehorter et al. 2011), a transient and significant increase in the expression of NR2A/B subunit-containing NMDA receptors in the second postnatal week and a gradual increase in AMPA/kainate receptor expression. The expression of UBP-310-sensitive kainate receptors seems to also progressively increase, whereas their contribution to the EPSP duration and decay time kinetics at later stages of development seems to decrease, suggesting complex developmental changes in glutamate receptor expression (Wisden & Seeburg, 1993;Bahn et al. 1994;Bischoff et al. 1997). The different subunits that make up the NMDA receptors and kainate receptors affect their channel kinetics (Monyer et al. 1994;Flint et al. 1997;Cull-Candy & Leszkiewicz, 2004;Chen et al. 2006;Lerma & Marques, 2013) and, in combination with changes in the membrane time constant (Spruston et al. 1994), might well contribute to the long duration EPSPs seen in the second postnatal week. Indeed, the observation of similar developmental changes in EPSC duration and decay time suggest that the observed effects cannot be explained solely by developmental changes in the membrane time constant. These long duration EPSPs could well play a role in facilitating synaptic integration and synaptic plasticity (Carmignoto & Vicini, 1992;Tang et al. 1999;Frerking & Ohliger-Frerking, 2002;Fino et al. 2010) at excitatory synapses on SPNs during this period and the generation of ensemble activity (Carrillo-Reid et al. 2008). Finally, the observation of a relatively stable mEPSC frequency was surprising in light of reported increases in the density of asymmetric glutamatergic synapses during this period (Butler et al. 1998;Tepper et al. 1998) and observed increases in dendritic spine density and previous reported increases in EPSC frequency (Peixoto et al. 2016). Possible explanations for our observations but also others (Dehorter et al. 2011) could include biases towards recording mEPSCs mediated by axo-dendritic synapses instead of axo-spinous synapses, as the density of axo-dendritic synapses have been shown to remain stable during development , or could result from a dynamic interplay between synapse number and release probability in which increases in synapse number are balanced by decreases in release probability (Choi & Lovinger, 1997).

Local inhibitory synaptic connections between D1 and D2 SPNs
The activity of striatal SPNs and their response to excitatory input is modulated by inhibition coming from local collaterals from neighbouring SPNs (Somogyi et al. 1981;Bolam & Izzo, 1988;Taverna et al. 2008;Planert et al. 2010;Cepeda et al. 2013) and striatal interneurons (Tepper & Plenz, 2006;Ponzi & Wickens, 2010). Measurements of mIPSCs reveal that both D1 and D2 SPNs already receive some inhibitory input in the first postnatal week. The initial frequency of these events is lower than that seen for mEPSCs and progressively increases in the second and later postnatal weeks as previously observed (Dehorter et al. 2011). This would suggest that GABAergic synapse density increases well after the second postnatal week. We also observed a dramatic increase in mIPSC amplitude from the first to second postnatal week consistent with a rapid maturation as a result of an increased number (Nusser et al. 1997) and/or changed subunit composition of postsynaptic GABA receptors (Farrant & Nusser, 2005;Arama et al. 2015). To investigate when and how SPNs start communicating with each other in the striatum and whether the number of synaptic connections does increase we performed simultaneous quadruple whole-cell patch-clamp recordings of SPNs. We found that both D1 and D2 SPNs are mainly connected through gap junctions in the first postnatal week, but both the incidence of gap junctions and their coupling coefficient rapidly decrease and no gap junctions were observed in adulthood, consistent with previous electrophysiological (Venance et al. 2004;Yu et al. 2012) and dye coupling experiments ). These initial gap junctions could facilitate synchronization of SPN activity (Venance et al. 2004;Hestrin & Galarreta, 2005) and the establishment of synaptic connections (Yu et al. 2012). The first inhibitory synaptic connections were detected in the first postnatal week coming from D1 SPNs only, potentially a reflection of their earlier birthdate (Marchand & Lajoie, 1986;van der Kooy & Fishell, 1987;Kelly et al. 2018), but in J Physiol 597.21 the second postnatal week both D1 and D2 SPNs form inhibitory synaptic connections with each other. Interestingly, the relative biases in connectivity (e.g. the high interconnectivity between D2 SPNs) seen in adulthood in this study and by others (Taverna et al. 2008;Planert et al. 2010) are already established by the second postnatal week raising the question of what instructs SPNs to form these precise intrastriatal circuit motifs (Plenz, 2003).
In conclusion, we show that early postnatal development of the intrinsic cellular and circuit properties of the D1-expressing direct pathway and the D2-expressing indirect pathway SPNs is highly dynamic but follows a clear developmental trajectory. Moreover, we show that many of the properties of mature D1 and D2 SPNs are already apparent by the first and second postnatal weeks, which is thought to be a period prior to much exploratory motor behaviour (Dehorter et al. 2011) or exposure to structured input from the sensory periphery (Tobach et al. 1971;Krug et al. 2001;Akerman et al. 2002;Ko et al. 2013;Mowery et al. 2015Mowery et al. , 2016Mowery et al. , 2017. This is consistent with the idea that neuronal specification can occur early in development (Lobo et al. 2006(Lobo et al. , 2008Arlotta et al. 2008;Ehrman et al. 2013;Lu et al. 2014;Zhang et al. 2016;Merchan-Sala et al. 2017;Kelly et al. 2018;Tinterri et al. 2018;Xu et al. 2018) with further postnatal development guided by, for example, neural activity (Zhang & Poo, 2001;Kozorovitskiy et al. 2012;Peixoto et al. 2016) and neuromodulation (Kozorovitskiy et al. 2015;Lieberman et al. 2018). Future work will be able to clarify the precise interaction of these factors during striatal development as well as their differential involvement in neurodevelopmental disorders (Graybiel & Rauch, 2000;Del Campo et al. 2011;Langen et al. 2011;McNaught & Mink, 2011;Shepherd, 2013;Albin, 2018).