Synapse-specific diversity of distinct postsynaptic GluN2 subtypes defines transmission strength in spinal lamina I

The unitary postsynaptic response to presynaptic quantal glutamate release is the fundamental basis of excitatory information transfer between neurons. The view, however, of individual glutamatergic synaptic connections in a population as homogenous, fixed-strength units of neural communication is becoming increasingly scrutinized. Here, we used minimal stimulation of individual glutamatergic afferent axons to evoke single synapse resolution postsynaptic responses from central sensory lamina I neurons in an ex vivo adult rat spinal slice preparation. We detected unitary events exhibiting a NMDA receptor component with distinct kinetic properties across synapses conferred by specific GluN2 subunit composition, indicative of GluN2 subtype-based postsynaptic heterogeneity. GluN2A, 2A and 2B, or 2B and 2D synaptic predominance functioned on distinct lamina I neuron types to narrowly, intermediately, or widely tune, respectively, the duration of evoked unitary depolarization events from resting membrane potential, which enabled individual synapses to grade differentially depolarizing steps during temporally patterned afferent input. Our results lead to a model wherein a core locus of proteomic complexity prevails at this central glutamatergic sensory synapse that involves distinct GluN2 subtype configurations. These findings have major implications for subthreshold integrative capacity and transmission strength in spinal lamina I and other CNS regions.


Introduction
The unitary postsynaptic response to quantal glutamate release is the fundamental basis of excitatory information transfer in the central nervous system (CNS). The prevailing assumption that glutamatergic synaptic connections in a given neural network or population function as homogenous, equal strength relay units of excitatory information is beginning to face critical re-evaluation (Grant and Fransén, 2020). The core components of a typical glutamatergic synapse include glutamate-gated ionotropic receptors (iGluRs) consisting primarily of α-amino-3-hydroxy-5-methyl-4-isoxazole propionate (AMPA) and N-methyl-D-aspartate glutamate receptor-channel (NMDAR) complexes, along with adhesion and signaling molecules Zhu et al., 2018;Cizeron et al., 2020). However, there is increased scrutiny directed toward a more comprehensive understanding of postsynaptic molecular composition and assembly-do a multitude of postsynaptic receptor-channel and protein constituents occur as a generic mix across synapses in a population, or is there a molecular codification to their nanopositioning and function that tunes transmission strength at individual glutamatergic synapses?
Recent advances in single-cell RNA sequencing and proteomics in the rodent brain (Roy et al., 2018a,b;Zhu et al., 2018;Cizeron et al., 2020) and spinal cord (Häring et al., 2018;Broadhead et al., 2020) have identified considerable diversity in synaptic RNA and protein content. At glutamatergic synapses in particular, molecular complexity is reflected extensively in the postsynaptic density (PSD) portion (Collins et al., 2006;Coba et al., 2009;Zhu et al., 2018;Broadhead et al., 2020;Cizeron et al., 2020) where AMPA and NMDA receptors are optimally situated for synaptic function. Though the NMDAR multiprotein structure itself is composed exclusively of ion channel subunits consisting of a heterotetramer core comprised of two obligate GluN1 subunits and two GluN2 subunits (Paoletti et al., 2013), NMDARs contained within "NMDAR supercomplexes" (Frank et al., 2016; are bound to a myriad of adhesion and signaling binding partners (Frank et al., 2016. Notably, dissimilar GluN2 subunits of which four types exist (2A-D) (Paoletti et al., 2013) are subjected to differing interactions with distinct PSD proteins (Kornau et al., 1995;Bassand et al., 1999;Valtschanoff et al., 1999;Losi et al., 2003;Lin et al., 2004;Gardoni and Di Luca, 2021) including for example PSD-95 (Cousins et al., 2008(Cousins et al., , 2009Bard et al., 2010;Cousins and Stephenson, 2012) indicative of GluN2 subtype regulation by distinct PSD binding partners. Although there is compelling evidence that NMDARs congregate in a cluster primarily at the center of the PSD (Kharazia and Weinberg, 1997;Chen et al., 2015;Goncalves et al., 2020), recent progress in vitro using singlemolecule and super-resolution imaging in cultured hippocampal neurons has now identified more precise spatial organization of distinct GluN2 subtypes in defined nanodomains in the PSD with the localization of each determined individually by specific PSD binding partners within the NMDAR supercomplex (Kellermayer et al., 2018;Hruska et al., 2022). The known extensive complexity of PSD protein constituents Roy et al., 2018b;Cizeron et al., 2020) and the regulatory function they impose on GluN2 subtype nanoscale localization in vitro (Kellermayer et al., 2018;Hruska et al., 2022) therefore open the potential for multiple scenarios of diverse GluN2 subtype(s) organization within the PSD compartment. However, our current understanding of GluN2 subtype diversity and distinct PSD compartmentalization at the resolution of individual synapses in higher-order mammalian synaptic circuits or populations remains elusive. Moreover, though GluN2 subunit composition within the NMDAR complex is known to determine several biophysical receptor-channel properties that mediate current flow (Clarke and Johnson, 2006;Siegler Retchless et al., 2012), the functional significance that distinct GluN2 subtype localization within the PSD imposes on unitary integrative capacity and transmission strength at individual synapses remains unexplored. Limited consideration may be dedicated to these issues likely because of the technically challenging nature of investigating NMDARs, their properties, and function at single-synapse resolution in in vivo or ex vivo CNS preparations.
We set out here to understand GluN2 subtype organization and function at the resolution of individual mature mammalian glutamatergic synaptic connections using an electrophysiological approach. We developed a minimal stimulation protocol in an ex vivo adult rat spinal cord slice preparation in which we evoked unitary excitatory postsynaptic currents (µEPSCs) and potentials (µEPSPs) at individual synaptic connections between glutamatergic primary afferents (Broman et al., 1993;Broman and Adahl, 1994;Todd et al., 2003) in the dorsal rootlet and central neurons in the lamina I nociceptive subfield of the spinal dorsal horn, a major CNS somatosensory region. µEPSCs exhibited distinctive NMDAR kinetic properties at this principal sensory synapse providing evidence supporting a high degree of diversity of single or multiple GluN2 subtype postsynaptic configurations at individual glutamatergic synaptic connections. µEPSP recordings further identified distinct GluN2 subtype configurations as a fundamental single-synapse resolution coding mechanism. Our results support a model wherein the mature mammalian primary afferent-lamina I neuron glutamatergic synapse is a core locus of distinct GluN2 subtype-based proteome diversity and nanoscale organization that directly influences fundamental aspects of subthreshold transmission efficacy including unitary resolution synaptic computation capability and strength, and ultimately how somatosensory information is processed. In other CNS regions diverse postsynaptic GluN2 subunit composition may constitute distinct integrative signatures at glutamatergic synapses that bear major implications in physiological transmission strength and information transfer.

Unitary primary afferent-lamina I neuron synaptic responses
To investigate NMDAR GluN2 subtypes at mature, lamina I synapses, we made whole-cell voltage-clamp recordings from visually identified lamina I neurons in acute ex vivo dorsal rootattached transverse spinal cord slices from adult rats. We elicited primary afferent-mediated post-synaptic responses in lamina I neurons by delivering single electrical stimuli to the attached dorsal rootlet. Stimulation of primary afferents synaptically connected to lamina I neurons evoked excitatory post-synaptic currents (EPSCs) (Figure 1A). When the membrane potential was held at hyperpolarized potentials, we observed a rapid-onset, rapidly decaying EPSC response. Holding the neuron at positive membrane potentials revealed a slowly decaying component of the EPSC, in addition to the rapid-onset peak. The peak of the rapid onset component exhibited a linear current-voltage relationship, whereas the slowly decaying EPSC component showed dramatic inward Primary afferent-evoked µEPSCs in whole cell voltage-clamp recordings from lamina I neurons in acute ex vivo dorsal root-attached transverse spinal cord slices from adult rats exhibit kinetic properties of both AMPA and NMDARs. (A) Delivery of a single high intensity electrical stimulus (gray arrow indicates stimulus artifact) to the attached rootlet that activates multiple primary afferent axons evokes large amplitude EPSCs in a lamina I neuron. Stimulation at hyperpolarized membrane potentials elicits rapid onset, rapidly decaying EPSC responses, whereas stimulation at depolarizing membrane potentials evokes a slowly decaying component, in addition to the rapid-onset peak. Peak of the rapid onset component shows a linear current-voltage relationship (inset below, white symbols) consistent with kinetic properties of AMPARs, while the slowly decaying EPSC component shows inward rectification (inset below, black symbols) consistent with current-voltage (I-V) relationship of NMDARs. (B) Recordings of unitary-evoked EPSCs from a representative adult rat spinal lamina I neuron elicited by minimal stimulation of an individual axon in the attached dorsal rootlet. Sample traces showing individual consecutive all-or-none µEPSCs held at + 60 mv evoked by minimal stimulation. The black traces show successful synaptic µEPSCs in response to minimal sensory afferent stimulation (gray arrow indicates stimulus artifact) while the gray trace shows a synaptic failure. (C) Scatter plot of successful synaptic µEPSC responses and failures over time (thick gray bar) evoked by minimal stimulation (0.2 Hz). The thin gray arrow indicates subthreshold stimulation intensity while the thick gray arrow indicates threshold at which µEPSCs can be evoked. (D) Several successful synaptic µEPSCs (black traces) from the same recording in panel (B) showing identical amplitude and latency (black arrows) consistent with a monosynaptic connection (gray arrow indicates stimulus artifact). The gray trace indicates a synaptic failure. (E) NMDAR amplitude histogram showing relative distribution for successful µEPSCs, fit with a single Gaussian function. (F) Averaged traces (± SEM) showing successful µEPSCs held at + 60 mV and from the same recording successful µEPSCs held at -65 mV. (G) Single exponential fitting of the decay component of the averaged µEPSC held at + 60 mV.
Frontiers in Synaptic Neuroscience 03 frontiersin.org rectification ( Figure 1A). The rapid-onset component of the EPSC was blocked by the AMPAR antagonist CNQX and the slow component was blocked by the NMDAR competitive antagonist D-amino-phosphonovaleric acid (D-APV, 100 µM; not illustrated). Thus, adult lamina I dorsal horn neurons exhibit primary afferentevoked EPSCs with kinetic and pharmacological properties of both AMPA and NMDA receptors.
To examine individual primary afferent-evoked EPSCs of lamina I neurons in the dorsal root-attached spinal cord slice preparation we used a minimal stimulation protocol (Hildebrand et al., 2014) to evoke unitary, monosynaptic EPSCs (µEPSCs) in the lamina I neuron recorded (n = 224 cells). The criteria used to establish monosynaptic µEPSC responses were: a distinct electrical stimulation threshold at which responses were all-ornone (i.e., success or failure; Figures 1B, C), a consistent latency throughout the duration of each recording (Figure 1D), and the amplitude of the NMDAR component was normally distributed (Figure 1E). We confirmed that responses evoked by the minimal stimulation were bona fide unitary responses because increasing stimulus intensity did not increase µEPSC amplitude until the intensity was sufficient to evoke a second level of all-or-none evoked responses (Supplementary Figure 1C). Moreover, bathapplying tetrodotoxin (TTX) to block voltage-gated Na + channels eliminated the µEPSCs (Supplementary Figures 1A, B), indicating that unitary synaptic responses were driven by action potentials evoked by direct stimulation of a presynaptic axon in the dorsal rootlet rather than by direct focal stimulation of the lamina I neuron.
To determine whether or not the response failures with minimal stimulation were due to failure of the stimulating electrode to generate an action potential in the primary afferent axon in the dorsal rootlet eliciting the µEPSC, we used paired-pulse minimal stimulation experiments (Supplementary Figures 1D, E). We found that the average amplitude of the µEPSCs to the second stimulus was greater than that of the first when the first stimulus also evoked a µEPSC (Supplementary Figures 1D, E, left panels), consistent with paired-pulse facilitation of transmitter release. Moreover, the average amplitude of the µEPSCs to the second stimulus that followed a failure of response to the first stimulus was identical to, if not greater than, the average of the responses to the second stimulus that followed first stimulus non-failures (Supplementary Figures 1D, E, right panels). These findings are not consistent with failure of the first stimulus to generate an action potential, in which case the amplitude of the µEPSC to the second stimulus would have been predicted to be the same as that of the first stimulus non-failures. Rather, our findings show that despite the failure of post-synaptic response to the first stimulus an action potential must have been triggered in the axon connected to the lamina I neuron, and must have reached the presynaptic terminal to enhance quantal release of glutamate presynaptically in response to the second stimulus. Therefore, we conclude that the minimal stimulation technique used here reliably evoked a monosynaptic, excitatory post-synaptic response from a single primary afferent connected to each lamina I neuron studied.
When initially searching for a unitary synaptic input in µEPSC recordings we held the membrane potential at −65 mv (n = 172 cells). In all of these recordings when the membrane potential was subsequently held at + 60 mv we observed a component of the decay phase that was slower than that when the neuron was held at −65 mv, which we interpret as indicating that there were both AMPAR-and NMDAR-components to the µEPSCs. In the remainder of recordings we searched with the membrane potential held at + 60 mv to look for NMDAR-only synapses. However, in all cases we found a rapid inward current when the neuron was subsequently held at −65 mv. Thus, we conclude that all single, primary afferent-to-lamina I neuron synapses we recorded contained both AMPAR and NMDAR components.
Three distinct µEPSC types evoked across individual adult primary afferent-lamina I neuron synapses Unitary excitatory postsynaptic currents average amplitude and decay were found to be stable within each single synapse recording (Vh: + 60 mV) throughout the duration of the recording period (e.g., Figures 1B-D, F, G). However, we observed considerable heterogeneity in the µEPSC decay between single-synapse recordings (Figure 2A). To gain insight into the diversity in µEPSCs between recordings we performed principal component analysis (PCA) followed by hierarchical clustering (HC). For the PCA we examined several parameters including charge transfer, half-width, decay constant, peak amplitude, rise time, neuronal input resistance, and synaptic failure rate. Dimensionality reduction of the parameters with PCA (Figures 2B, C and Supplementary Figure 2), showed that the first three components capture 76% of the data variability. Based on this, µEPSC decay, charge transfer and half-width were retained as electrophysiological parameters that significantly contributed to the first three PCA components ( Figure 2C) and, therefore, used to build a classification of individual synaptic responses. To determine whether primary afferent-lamina I neuron synapses are comprised of distinct types, we used µEPSC decay, charge transfer and half-width for the HC ( Figure 2D). Gap-statistic analysis of the clustering ( Figure 2D inset top right) identified three as the optimal number of clusters to represent the dataset. We labeled these three groups: fast (n = 80, 36% of recorded neurons), intermediate (n = 114, 51% of recorded neurons), and slow (n = 30, 13% of recorded neurons) based on decay time constant of their µEPSCs ( Figure 2E). Individual representative fast, intermediate, and slow µEPSCs are shown in Supplementary Figure 3. From these representative µEPSCs, latency, amplitude, and decay can be seen to be stable throughout each recording (Supplementary Figure 4). Mean µEPSC traces for each of the three unitary synaptic response types determined by clustering are shown in Figures 3A-C. Across these three groups the following electrophysiological parameters were not different: failure rate (fast 56 ± 2%, intermediate 52 ± 2%, and slow 51 ± 3%), cell input resistance (fast 363 ± 32 MOhm, intermediate 334 ± 19 MOhm, and slow 367 ± 34 MOhm) as well as µEPSC rise time (fast 2.9 ± 0.2 ms, intermediate 3.1 ± 0.2 ms, and slow 3.4 ± 0.3 ms). Therefore, we conclude that primary afferent-lamina I µEPSCs at the holding potential of + 60 mv can be classified into three major groups according to decay constant, half-width, and charge transfer ( Figure 3D).
Across these three groups, the µEPSCs at the holding potential of −65 mv did not differ in peak amplitudes (fast −24 ± 1.4 pA, intermediate −26 ± 0.9 pA, and slow −24 ± 1.6 pA; p > 0.05) or  Figure 3E). Moreover, the peak amplitudes of µEPSCs held at −65 mv exhibited linear current-voltage relationships ( Figure 3F). Thus, the AMPAR components of the µEPSCs were indistinguishable across the three groups. Distinct GluN2 subtypes mediate the decay kinetics of the three distinct unitary synaptic responses As the NMDAR component but not the AMPAR component of the uEPSCs fell into three groups we investigated the basis for the differences in the NMDAR component. Of the three kinetic parameters that account for variance within the population of µEPSCs, half-width and charge transfer (calculated as area under the curve) occur as a passive function of the time course of the µEPSC decay (Figure 3). Thus, we inferred from the readily discernible correlation of charge transfer and half-width with decay that the most critical parameter governing the functional importance of each unitary response is time course of the decay itself. We therefore focused our attention on elucidating the mechanism(s) responsible for different µEPSC decay kinetics. Notably, given that the three distinct µEPSC decay constants determined from our cluster analysis (52 ± 3 ms for "fast" decay; 315 ± 11 ms for "intermediate" decay; 1,219 ± 69 ms for "slow" decay at Vh: + 60 mV; p < 0.001; Figure 3D) correspond well with values from known NMDAR GluN2 subtypes including specifically GluN1/2A/2A, 2B/2B, and 2D/2D, respectively (Cull-Candy and Leszkiewicz, 2004;Hildebrand et al., 2014), we hypothesized that the dissimilar, yet D-APV-sensitive (Supplementary Figure 5), decay components of unitary synaptic responses represent distinct deactivation kinetics governed by differences in intrinsic properties of distinct GluN2 subtype-containing NMDARs at these synapses.
To test this hypothesis, we rigorously investigated the pharmacological sensitivity of the three distinct response types to known selective NMDAR GluN2 subtype inhibitors. We tested first involvement of GluN2D/2D-containing NMDARs. In a subset of voltage-clamp recordings examining representative "fast, " "intermediate, " and "slow" µEPSC responses (Vh: + 60 mv), we administered to aCSF the GluN2D/2D selective antagonist, DQP-1105 (10 µM), which is selective for GluN2D/2D diheteromeric NMDARs (Acker et al., 2011) which we refer to here as GluN2D.  Supplementary Figures 6D, E). Taken together, these results indicate that the particularly prolonged decay of the "slow" µEPSC response was mediated by GluN2Dcontaining NMDARs.
We investigated next possible involvement of GluN2B/2Bcontaining NMDARs by bath administering the GluN2B/2B selective antagonist, Ro25-6981 (1 µM) which is selective for GluN2B/2B diheteromeric NMDARs (Fischer et al., 1997) which we refer to here as GluN2B. Ro25-6981 was without effect on "fast" decay µEPSC responses (Supplementary Figures 7C, D . We therefore conclude that both "intermediate" and "slow" synapse types are comprised, at least in part, of GluN2B-containing NMDARs. Having shown that the NMDAR component of "fast" µEPSC responses was resistant to both Ro25-6981 and DQP-1105 (Supplementary Figures 6, 7), we reasoned that "fast" unitary primary afferent-lamina I synaptic responses are mediated by NMDARs other than GluN2B-or GluN2D-containing subtypes. We considered, therefore, the possibility that GluN2A/2Acontaining diheteromeric NMDARs which are known to exhibit decay kinetics faster than those of the other GluN2 subunits (Hildebrand et al., 2014) may underlie this NMDAR-mediated synaptic response. To investigate potential function of GluN2A/2Acontaining NMDARs (which we refer to here as GluN2A) at these synapses we administered the selective antagonist, TCN-201 (3 µM) (Hansen et al., 2012). We tested TCN-201 on representative "fast" decay µEPSC responses and found that the µEPSC duration was reduced (Supplementary Figures 8A-C) to the extent that the entire charge transfer mediated by NMDAR function was abolished (p < 0.01; Supplementary Figures 8D, E). We conclude that GluN2A-containing NMDARs comprise the "fast" synapse type.
The depressive effect of Ro25-6981 on slow decay µEPSCs (Supplementary Figures 7A, D, E) suggests function of GluN2Bcontaining in addition to GluN2D-containing NMDARs at "slow" decay synapses. In addition, that the residual "DQP-1105insensitive" NMDAR component of "slow" decay µEPSCs showing sensitivity to DQP-1105 is comparable to "intermediate" µEPSC responses (Supplementary Figures 6A, D-F) further provides evidence for GluN2B subtype NMDARs in addition to GluN2Dcontaining NMDARs at "slow" decay synapses. Furthermore, though we did not examine specifically TCN-201 on "intermediate" and "slow" decay µEPSC responses, we are not ruling out a contribution of GluN2A-containing NMDAR function at these synapses. For example, the τ (fast) for "intermediate" decay µEPSCs (45 ± 4 ms) is comparable (p > 0.05) to the τ value for "fast" (GluN2A-mediated) µEPSC decay kinetics (52 ± 3 ms; Figure 3D). It is also evident from "intermediate" decay µEPSCs which showed sensitivity to Ro25-6981 that the remaining "Ro25-6981-insensitive" NMDAR component is comparable to the "fast" decay µEPSC responses (Supplementary Figures 7C, D, F, G). Based on these observations, we are not excluding the possibility that "intermediate" and possibly "slow" synapses may also contain GluN2A-containing NMDARs.
Differential Mg 2+ block of fast, intermediate, and slow µEPSCs GluN2 subunit differences in the kinetics of Mg 2+ unblock contribute to NMDAR subtype function in current flow during membrane depolarization at glutamatergic synapses (Clarke and Johnson, 2006). Given the distinct GluN2 kinetic properties determined from the three µEPSC types recorded at the holding potential of + 60 mv ( Figures 3A-D), we reasoned that µEPSCs recorded from the same experiments but at the negative holding potential of −65 mv (Figures 3A-C bottom) would also reflect differences in biophysical properties indicative of distinct GluN2 subtypes. Consistent with this prediction, the late component of negatively held µEPSCs exhibited significant differences in halfwidth (p < 0.01) and charge transfer (p < 0.05, 0.01, 0.001) across the three groups (Supplementary Figure 9A) that may be attributed to differences in sensitivity of GluN2 subtypes to Mg 2+ block (Kuner and Schoepfer, 1996;Clarke and Johnson, 2006). By determining the Mg 2+ -induced percent inhibition of the amplitude of the NMDAR component of µEPSCs recorded at the holding potential of −65 mv in each group (Supplementary Figure 9B) we found that the inhibition of the NMDAR component of µEPSCs from the "fast" cluster was similar to that of µEPSCs from the "intermediate" cluster, whereas the Mg 2+ -induced percent inhibition of the NMDAR component of µEPSCs from the "slow" cluster was substantially less (Supplementary Figure 9B; Kuner and Schoepfer, 1996;Clarke and Johnson, 2006). We conclude that the differences in Mg 2+ block of µEPSC NMDAR components at primary afferent-lamina I glutamatergic synapses further confirm the distinctiveness of GluN2-containing NMDAR subtype expression and function we propose to occur at these synapses.

No evidence for presynaptic NMDAR function
We reasoned that if NMDARs functioned presynaptically to influence synaptic transmission, we would observe a change in unitary synaptic failure rate. We found that failure rate remained stable in recordings in which D-APV, TCN-201, Ro25-6981, or DQP-1105 was tested (Supplementary Figures 10A-C). In a representative experiment, D-APV is shown to inhibit the NMDAR component of the µEPSC (mean ± SEM) while exhibiting no effect on failure rate (Supplementary Figure 10A) consistent with a postsynaptic function of NMDARs at primary afferent-lamina I neuron synapses.

µEPSP temporal summation at individual synapses is NMDAR-dependent
The diversity of GluN2-containing NMDARs at individual lamina I synapses and their differences in Mg 2+ block prompted us to investigate next the function of GluN2 subtype diversity at individual primary afferent-lamina I neuron synapses. We carried out minimal stimulation in voltage-clamp first to establish bona fide µEPSC responses (Vh: −65 mv) after which, within the same recording, we switched to current-clamp mode to monitor membrane potential (µEPSP response using a K + -based intracellular solution). In this dual mode recording protocol we used a brief train composed of 5 stimuli delivered every 200 ms [5 Hz frequency which mimics low repetitive firing pattern of nociceptors (Matsumoto et al., 2006)] at 5 s intervals (0.2 Hz) in voltage clamp (Supplementary Figures 11A, B) and then current clamp (Supplementary Figures 11C, D) modes for a given lamina I recording.
Continuous sweep recordings of µEPSCs during the 5 Hz minimal stimulus train show that mean (± SEM) peak amplitude, half-width, and rise time remained stable (p > 0.05; Supplementary  Figures 11A, B). Though an increase in µEPSC mean (± SEM) charge transfer was observed over successive stimuli (p < 0.01), pure µEPSC responses (without failures) at each stimulus throughout the 5 Hz train remained stable; peak amplitude, charge transfer, half-width, and rise time for the five µEPSCs in the 5 Hz train were not significantly different (p > 0.05; Supplementary  Figures 12A, B). A decrease in failure rate was found between the first and fifth pure µEPSCs (successes only) in the 5 Hz train (p < 0.05; Supplementary Figure 12B). Though indicative of increased presynaptic glutamate release probability through the progression of the 5 Hz train, we do not consider this to have influenced postsynaptic mechanisms during the train because each pure µEPSC response (without failures) remained constant.
Continuous sweeps recorded in current clamp mode during the 5 Hz train showed that membrane potential depolarized progressively with each stimulus, peaking after the fifth stimulus, a response we refer to as a "µEPSP burst" (Supplementary  Figures 11C, D). Pure µEPSP responses (successes only) shown at each stimulus (mean ± SEM) during the 5 Hz train exhibited a progressive increase in peak amplitude, area, half-width, decay, and prestimulus membrane potential (p < 0.001; Supplementary  Figures 13A, B), while rise time across the five µEPSPs in the train was not different (p > 0.05; Supplementary Figures 13A,  B). Notably, the failure rates for the first and fifth µEPSPs were indistinguishable from those for the corresponding first and fifth µEPSCs recorded in voltage clamp (p > 0.05; Supplementary Figure 12B) which indicate that the transition in recording mode (from voltage clamp) was inconsequential to mechanisms underlying unitary synaptic transmission as we were able to evoke reliable µEPSPs that summate with a failure rate that was indiscernible from that during µEPSC recordings.
In a representative example µEPSP burst, D-APV was without effect on the rising phase or the peak amplitude of the first µEPSP in the 5 Hz train but reduced µEPSP decay and abolished temporal summation (Supplementary Figure 14A). Similarly, the NMDAR channel blocker, MK-801, reversed the summated component of the µEPSP burst (Supplementary Figure 14B). Taken together, we conclude that temporally patterned µEPSPs evoked during the 5 Hz stimulus train exhibit summation and that µEPSP decay and summation are dependent on NMDAR function.

Distinct µEPSP kinetics across individual primary afferent-lamina I neuron synapses
Having determined that that temporal summation of µEPSPs at this synapse is attributable to NMDAR function (Supplementary Figure 14), we reasoned that GluN2 subtype synaptic distinctiveness (Figures 2, 3 and Supplementary  Figures 2-9) may underlie possible heterogeneity in our µEPSP burst dataset. In one approach to elucidate potential diversity in µEPSP burst properties, we grouped µEPSP bursts based on lamina I neuron type (Figures 4A-C). In each current-clamp recording, in which we had acquired also membrane potential in response to depolarizing current injection (20 pA for 1 s), we identified in lamina I neurons distinct firing properties in response to current injection. This enabled us a subclassification of our µEPSP dataset based on neuron spiking properties including single/non-spiking, delayed/phasic spiking, or tonic spiking activity (Figures 4A-C), established features of lamina I neurons (Yasaka et al., 2010;Punnakkal et al., 2014;Browne et al., 2020). We examined kinetic properties from the first µEPSP in the burst as it was the µEPSP in the stimulus train without summation, and from the last (fifth) µEPSP in the burst as it exhibited the greatest degree of summation (Supplementary Figure 13). Within lamina I neuron type, pure µEPSPs (successes only, mean ± SEM) exhibited an increase in area, half-width, decay, and prestimulus membrane potential of the fifth µEPSP vs. the first µEPSP (p < 0.05, p < 0.01, p < 0.001, Figures 4E-H). Between lamina I neuron type, pure first and fifth µEPSPs exhibited differences in area, half-width, and decay (p < 0.05, p < 0.01, p < 0.001; Figures 4E-G). Pure fifth µEPSPs between neuron type also exhibited significant differences in peak amplitude and prestimulus membrane potential (p < 0.05, p < 0.01, p < 0.001; Figures 4D, H). These results show that primary afferent-lamina I neuron synapses possess distinct kinetic properties that enable µEPSP temporal summation capability (Figures 4D-H). Importantly, µEPSP summation capacity at a given primary afferent synapse shows correlation with lamina I neuron type (single/non-spiking < delayed/phasic spiking < tonic spiking).
In another approach to elucidate potential µEPSP diversity we applied unsupervised HC analysis to the µEPSP burst dataset. µEPSP kinetic properties including pure first and fifth µEPSP half-width, area, and decay were used for HC which identified three distinct clusters (Figures 5A-C) referred to as "fast, " "intermediate" and "slow" kinetic properties which we compared to µEPSP classification based on lamina I neuron type (Figure 5D). Clusters representing first and fifth µEPSP responses from tonic spiking neurons (Figure 5D) corresponded entirely with clusters representing "slow" µEPSP kinetics for both the first and fifth First and fifth µEPSP responses recorded from single/non-spiking, delayed/phasic spiking, and tonic spiking lamina I neuron types. (A) Pure mean (± SEM) µEPSP traces (without synaptic failures) shown for the first (A) and fifth (B) minimal stimulus during the 5 Hz train (arrows indicate stimulus artifacts) for single/non-spiking, delayed/phasic spiking, and tonic spiking lamina I neurons. (C) Representative responses to current injection (20 pA for 1s) of each lamina I neuron type including single/non-spiking (inset at right shows magnified view), delayed/phasic spiking, and tonic spiking. (D) Scatter plots (mean ± SEM) showing µEPSP peak amplitude for the first and fifth µEPSPs from tonic spiking (black symbol), delayed/phasic spiking (gray symbol), and single/non-spiking (white symbol) lamina I neurons (µEPSP #1 vs. #5, * p < 0.05, ** p < 0.01, or *** p < 0.001, paired t-test; + p < 0.05, + + p < 0.01, or + + + p < 0.001 between lamina I spiking types at µEPSP #1 and #5, t-test), (E) area, (F) half-width, (G) decay, and (H) prestimulus membrane potential.
µEPSP responses (Figures 5C-E, black symbols) indicative of a homogenous distribution of primary afferent synapses with "slow" µEPSP kinetic properties occurring on tonic spiking lamina I neurons ( Figure 5E). Clusters representing delayed/phasic spiking neurons ( Figure 5D, gray symbols) were affiliated predominantly with "intermediate" clusters for the first and fifth µEPSPs ( Figures 5C-E) while a small portion of µEPSP responses were affiliated with either "fast" (Figures 5C, E) or "slow" (Figures 5C,  E) clusters. Clusters representing single/non-spiking neurons for the first and fifth µEPSPs ( Figure 5D, white symbols) overlapped with both "fast" and "intermediate" kinetics (Figures 5C-E, white and gray symbols) but not with "slow" kinetics (Figures 5C-E, black symbols). We conclude that first and fifth µEPSPs can be classified into three major groups consisting of "fast, " "intermediate, " or "slow" kinetics according to half-width, area, and decay (Figures 5A-C). Notably, single/non-spiking lamina I neurons contain primary afferent synapses with predominantly "fast" and "intermediate" postsynaptic kinetic properties, while delayed/phasic spiking neurons contain primary afferent synapses with predominantly "intermediate" postsynaptic properties and some with "fast" kinetic properties. Tonic spiking neurons contain primary afferent synapses with exclusively "slow" postsynaptic kinetic properties.
Link between lamina I neuron type and µEPSP summation capacity is GluN2 subtype-dependent Having determined that primary afferent-lamina I neuron synapses contain distinct NMDAR GluN2-containing subtypes (Figures 2-3 and Supplementary Figures 2-9), that temporal summation of µEPSPs at these synapses is attributable to NMDAR function (Supplementary Figure 14), and that HC identified three distinct µEPSP burst clusters, we reasoned that intrinsic properties of µEPSP burst summation capacity derive from distinct primary afferent-lamina I neuron postsynaptic GluN2 subtype-containing NMDARs. To elucidate the role of GluN2 subtype in µEPSP temporal summation, we therefore probed the pharmacological sensitivity of primary afferent-evoked µEPSP bursts recorded from lamina I neurons with known spiking patterns using the selective inhibitor for GluN2A (TCN-201; 3 µM), GluN2B (Ro25-6981; 1 µM), or GluN2D (DQP-1105; 10 µM) administered to aCSF. Baseline µEPSP burst area (mean ± SEM) determined from single/non-spiking (88 ± 16 mv * ms), delayed/phasic spiking (115 ± 20 mv * ms), and tonic spiking (295 mv * ms) neurons (Figures 6D, H, K, respectively) exhibited a progressive increase across neuron type consistent with neuron type-specific temporal summation in our dataset ( Figure 4E). TCN-201 inhibited µEPSP burst responses from single/non- (Figures 6A, D; n = 4) and delayed/phasic spiking lamina I neuron (Figures 6F, H; n = 1) but was without effect on a µEPSP burst recorded from a tonic spiking lamina I neuron (Figures 6I, K; n = 1). Ro25-6981 was also found to inhibit µEPSP burst responses from single/nonspiking (Figures 6C, D; n = 1) and delayed/phasic spiking (Figures 6E, H; n = 4) lamina I neurons. However, in experiments in which we tested involvement of GluN2D-containing NMDARs, DQP-1105 was without effect on µEPSP bursts recorded from single/non-spiking (Figures 6B, D; n = 1) and delayed/phasic spiking (Figures 6G, H; n = 3) lamina I neuron types but inhibited the µEPSP burst recorded from a tonic spiking lamina I neuron (Figures 6J, K; n = 1). Our results identify GluN2A and 2B subtype function at primary afferent synapses on single/nonspiking and delayed/phasic spiking lamina I neurons while the GluN2D NMDAR subtype functions predominantly at primary afferent synapses on tonic spiking lamina I neurons. Taken together, we conclude that GluN2 subtypes occur with diversity across different lamina I neuron types and that µEPSP summation at the level of individual primary afferent-lamina I neuron synapses is dependent on distinct GluN2 subtype(s) composition and function at the synapse.

Discussion
Our strategy to examine single synapse fidelity postsynaptic responses to evoked primary afferent input in the adult rat spinal lamina I subfield identifies this sensory synapse as a major locus of diversity that arises from a heterogenous postsynaptic compartmentalization of single or multiple distinct GluN2 subtypes. We show further that the distinct postsynaptic GluN2 subtype(s) configuration at a given individual afferent synapse on a lamina I neuron defines precisely the computational capability and strength of that synapse.

Single or multiple GluN2 subtype configurations at primary afferent-lamina I neuron synapses
Principal component and hierarchical analysis combined with GluN2 pharmacological testing of µEPSCs recorded from hundreds of individual primary afferent-lamina I neuron synapses identified distinct postsynaptic GluN2 subtype configurations. To our knowledge this is the first demonstration of higherorder assembly of distinct GluN2 subtypes of the NMDAR occurring either individually or in defined combinations at a fully developed, mammalian, principal sensory glutamatergic synapse. Super-resolution imaging in cultured hippocampal neurons has described precise control of the spatial organization of distinct GluN2 subtypes (GluN2A and 2B to date) in defined postsynaptic nanodomains in the PSD by specific PSD protein constituents (Kellermayer et al., 2018;Hruska et al., 2022). In the present study, minimal stimulation in our ex vivo slice preparation enabled us to evoke glutamate release at single synapse resolution and therefore identify at individual synapses functionally relevant postsynaptic GluN2 localization(s) which we found to consist of GluN2A alone, GluN2A and 2B, or GluN2B and 2D. We consider these findings as conclusive evidence of functionally distinct postsynaptic GluN2 subtype organization at individual mammalian glutamatergic synapses.
Previous investigations documenting GluN2 subtype content in spinal lamina I and II have used high intensity stimulationevoked EPSCs (Tong et al., 2008;Shiokawa et al., 2010;Tong and MacDermott, 2014) or spontaneously occurring mEPSCs (Hildebrand et al., 2014) from populations of synapses. Although these studies provide insight into GluN2 subtype content in the superficial dorsal horn, stimulus protocols that elicit activation of a synapse population are not amenable to elucidating single or multiple GluN2 subtype configurations and function at an individual primary afferent-lamina I neuron synapse. Furthermore, recording miniature or spontaneous unitary synaptic events do not resolve responses in a population of synaptic inputs from responses of an individual synapse. The single-synapse fidelity of our recordings here enabled us to acquire and study defined unitary synaptic events at individual primary afferent-lamina I neuron monosynaptic connections and investigate at high resolution GluN2 subtype diversity and function.

Functional implications of GluN2 subtype on synaptic computation and transmission in spinal lamina I
Voltage-dependent inward currents mediated by sodium and calcium channels have been proposed to modulate EPSP pattern and integration in the brain (González-Burgos and Barrionuevo, 2001) and spinal cord dorsal horn (Prescott and De Koninck, 2005). The differences in Mg 2+ block of the distinct µEPSC NMDAR components found here at individual primary afferentlamina I neuron glutamatergic synapses (Supplementary Figure 9) not only substantiate the postsynaptic organization of GluN2A, GluN2A and 2B, or GluN2B and 2D we propose to occur at these synapses but highlight their distinct biophysical properties with implications in current flow through NMDARs during membrane depolarization (Kuner and Schoepfer, 1996;Clarke and Johnson, 2006). We show here that evoked unitary depolarizations at primary afferent-lamina I neuron monosynaptic connections are dependent on both AMPAR-and NMDARs. Our findings further indicate that µEPSP decay and summation throughout the µEPSP burst is shaped primarily by NMDAR-mediated current (Figure 6 and Supplementary Figure 14) opening the possibility that NMDARs also contribute to the integrative kinetics underlying subthreshold synaptic membrane depolarization (Pitcher et al., 2011).
In addition to acquiring evoked µEPSP bursts, current injection in our recordings elucidated spiking characteristics of lamina I neurons including single/non-spiking, delayed/phasic spiking, or tonic spiking (Yasaka et al., 2010;Punnakkal et al., 2014;Browne et al., 2020). This enabled us to classify µEPSP burst properties by neuron spiking type (Figure 4). However, a definitive correlation could not be made with the three µEPSP clusters categorized by lamina I neuron spiking type and the three clusters of µEPSP kinetics determined by HC analysis (Figures 5C-E). Though µEPSPs clustered by the tonic spiking lamina I neuron type corresponded entirely with the slower and more prolonged µEPSPs classified by HC analysis, this was not the case for the other two neuron types. Delayed/phasic spiking lamina I neurons exhibited µEPSP kinetic properties that corresponded predominantly with those of intermediate µEPSP kinetics classified by HC while a small percentage of µEPSPs exhibited kinetic properties consistent with fast and slow kinetics defined by HC. Furthermore, µEPSPs evoked at single/non-spiking lamina I neurons exhibited kinetic properties that corresponded almost equally with fast and intermediate µEPSP kinetic properties determined by HC. The overall concept that emerges from these results is that µEPSP classification is not exclusively neuron specific but rather synapse specific in that µEPSP kinetics in a given HC-defined group can be found in different lamina I neuron types. µEPSP kinetic properties that are either identical for a given neuron type, i.e., neuronspecific, or distinctly different within a given neuron type, i.e., synapse-specific, exemplify diversity in the individual µEPSP itself as well as temporal summation capability at primary afferentlamina I neuron synapses that have major functional implications in subthreshold synaptic computation and strength for lamina I neurons to temporally patterned sensory input. No other study that we are aware of has demonstrated this degree of heterogeneity in primary afferent-evoked µEPSP responses at individual synapses within and between distinct lamina I neuron types.
The three µEPSP clusters we report here are unexpected because they do not correspond to unitary synaptic depolarization kinetics reported previously in the spinal dorsal horn (Prescott and De Koninck, 2005). Prescott and De Koninck (2005) found that subthreshold depolarization measured from spontaneous EPSPs (sEPSPs) was similar to that from both delayed/phasic and single spiking lamina I neurons while sEPSPs from tonically spiking lamina I neurons exhibited markedly prolonged depolarization kinetics (Prescott and De Koninck, 2005), consistent with two classifications of unitary sEPSP kinetics. We do not have a clear explanation for this apparent discrepancy, but the possibility can't be excluded that the spontaneous events in the study by Prescott and De Koninck (2005) may have derived from synaptic inputs other than primary afferents. We therefore consider the prospect that mechanisms underlying the kinetics of unitary depolarizations at the primary afferent-lamina I synaptic connection may be distinctly different from those at nonprimary afferent synaptic connections. In addition, the number of unitary synaptic recordings obtained in the present study also merit consideration in that they are relatively high and may therefore favor increased opportunity to detect and decipher previously unrecognized synaptic distinctiveness of NMDARdependent biophysical properties of evoked µEPSPs at the primary afferent-lamina I neuron synapse.
The three µEPSC and three µEPSP clusters identified by HC analysis (Figures 2, 5, respectively) support the idea that three distinct GluN2-dependent mechanisms may be at play in mediating different stimulus-evoked membrane depolarization at individual primary afferent-lamina I neuron synapses. The selective GluN2D inhibitor, DQP-1105, was found to be without effect on fast and intermediate µEPSP decay and burst responses in recordings from single/non-spiking and delayed/phasic lamina I neurons (Figures 6B, D, G, H), but depressed the µEPSP burst exhibiting slow decay kinetics and robust temporal summation recorded from a tonic spiking neuron (Figures 6J, K). Thus, GluN2D subtype-dependent slow kinetic properties therefore function to broaden µEPSP decay and integrative capacity, and ultimately facilitate burst summation capability to temporally patterned afferent synaptic input to tonic spiking lamina I neurons. The selective GluN2A inhibitor, TCN-201, was without effect on slow µEPSP decay and summation in a recording from a tonic spiking lamina I neuron (Figures 6I, K), but depressed µEPSP bursts exhibiting fast decay kinetics and low temporal summation in recordings from single/non-and delayed/phasic spiking neurons (Figures 6A, D, F, H). GluN2A subtype-dependent fast kinetic properties therefore function to accelerate µEPSP decay and limit integrative capacity to temporally patterned afferent synaptic input. The selective GluN2B inhibitor, Ro25-6981, depressed µEPSP bursts displaying intermediate kinetic properties in recordings from single/non-and delayed/phasic spiking lamina I neurons (Figures 6C-E, H) indicative of GluN2B subtype-dependent function in mediating intermediate µEPSP decay and temporal summation. The main concept that emerges from our data is that decay and summation capacity of µEPSPs evoked at a given primary afferent-lamina I neuron synapse are functionally dependent and shaped by the distinct GluN2 subtype synaptic predominance at that synapse which signals to narrowly, intermediately, or widely tune, respectively, the duration of evoked unitary depolarization from resting membrane potential. Ultimately, this would have major implications in subthreshold integrative capacity of different lamina I neuron types to temporally patterned glutamatergic input from an ensemble of primary afferents.
The effects of the selective GluN2 subtype inhibitors examined in our µEPSC recordings revealed that the GluN2 subtype with the longest decay at a given synapse is the main identifiable GluN2 subtype responsible for carrying the greatest charge transfer at that synapse (Supplementary Figures 6, 7). Implicit to these results is that the GluN2 subtype(s) exhibiting residual kinetics identifies multiple possible GluN2 subtype combinations localized at a synapse. Taking into account the GluN2A, 2A and 2B, and 2B and 2D synaptic configurations determined in our voltage clamp experiments, we therefore propose the scenario in which primary afferent synaptic connections to single/non-spiking lamina I neurons may contain GluN2A only or GluN2A and 2B postsynaptically, while primary afferent synaptic connections to delayed/phasic neurons may contain postsynaptically GluN2A and 2B and some GluN2A only. Primary afferent synapses on tonic spiking lamina I neurons appear to be comprised exclusively of GluN2B and 2D postsynaptically. Although these data together provide evidence to suggest postsynaptic localization of distinct single or multiple GluN2 subtypes at individual synapses on specific lamina I neuron types, it is unclear the functional necessity or significance of multiple distinct GluN2 subtypes exhibiting differing kinetic properties at a given synapse. We reason that multiple postsynaptic GluN2 subtype configurations may endow primary afferent-lamina I neuron synapses with a more diverse GluN2 subtype-dependent modulation of membrane potential (and possibly other forms of GluN2 subtype-dependent signaling) that imparts stringent control of transmission efficacy.
Single cell transcriptional analysis of neurons in the rodent spinal dorsal horn (Häring et al., 2018) has identified 15 inhibitory (GABAergic) and 15 excitatory (glutamatergic) molecular subtypes of neuron in which Grin2b occurs abundantly in both excitatory and inhibitory subtypes, while Grin2d is low in excitatory and moderately expressed in inhibitory neurons, and Grin2a is found to be low in inhibitory and low to moderate in excitatory neurons. We did not explicitly attempt here to resolve GluN2 subtype function in inhibitory vs. excitatory lamina I neurons. We did, however, acquire the spiking pattern of the lamina I neurons recorded in current clamp, as described above. Notably, tonic spiking spinal dorsal horn neurons are generally classified as inhibitory (Yasaka et al., 2010;Punnakkal et al., 2014), but accumulating evidence shows a proportion of this neuron spiking type to be excitatory (Yasaka et al., 2010;Browne et al., 2020). Furthermore, though delayed/phasic spiking dorsal horn neurons have been shown to be inhibitory (Punnakkal et al., 2014;Browne et al., 2020), they have also been shown to be predominantly excitatory (Yasaka et al., 2010;Punnakkal et al., 2014;Browne et al., 2020). Similarly, though a small proportion of single/non-spiking dorsal horn neurons have been found to be inhibitory (Punnakkal et al., 2014;Browne et al., 2020), this neuron type has also been found to be predominantly excitatory (Yasaka et al., 2010;Punnakkal et al., 2014;Browne et al., 2020). Notwithstanding these apparent discrepancies, the synapse-specificity we propose here, rooted in distinct GluN2 subtype configurations and computation capability, may represent a basis on which multiple CNS mechanisms (e.g., excitatory and inhibitory) are driven by glutamatergic synaptic input and tightly modulated by distinct postsynaptic GluN2 subtype configurations.

Synaptic GluN2 subtype heterogeneity as a determinant of transmission fidelity in the CNS
We used the ex vivo primary afferent-lamina I neuron synaptic preparation as a model glutamatergic circuit in which single-synapse fidelity postsynaptic responses were investigated. Based on our results showing synapse-specific GluN2 subtype heterogeneity across individual glutamatergic synapses, we propose for consideration major roles of GluN2 subtype diversity in neuron excitability throughout the CNS. The functional dependence of µEPSP shape and µEPSP burst summation capacity on the diverse single or multiple GluN2 subtype configurations with differing kinetics at a given synapse would be predicted, in an ensemble of glutamatergic inputs, to impact EPSP-spike coupling (Li and Gulledge, 2021) by gauging membrane potential depolarization and, ultimately, the neuron reaching spiking threshold. We also put forward here the concept that GluN2 protein subtype heterogeneity in synaptic ensembles to a given neuron may serve a coding function that coordinates the synaptic integrative fidelity of that neuron to repetitive input activity (coincidence detection) which, importantly, would further influence neuron firing.
The NMDAR dendritic spike is the dominant mechanism by which dendritic integration of glutamatergic inputs through the recruitment of NMDAR channels leads to firing of a neuron (Schiller and Schiller, 2001;Antic et al., 2010). Since its discovery in cortical pyramidal cell dendrites over 20 years ago (Schiller et al., 2000) the NMDAR spike continues to receive considerable focus in different cortical regions (Enoki et al., 2004;Manita et al., 2011;Lavzin et al., 2012;Palmer et al., 2014;Kumar et al., 2018;Otor et al., 2022). Notably, recent evidence has uncovered critical features of NMDAR-containing synapses that contribute to the generation of spikes including synapse number (Plotkin et al., 2011), proximity to other synapses (Plotkin et al., 2011), spatial clustering (Poleg-Polsky, 2015Weber et al., 2016;Scholl et al., 2021), as well as their dendritic location (Froemke et al., 2010;Menon et al., 2013;Gao et al., 2021). Given our findings here, we propose for consideration the potential role GluN2-dependent synaptic diversity and organization may impose on dendritic spike threshold and efficacy in firing in neurons in which NMDAR spikes predominate.
Molecular and architectural complexity across glutamatergic synapses, arising from a multitude of differences in postsynaptic proteome constituents and their nanostructure organization Roy et al., 2018a,b), are beginning to emerge as determinants of individual synaptic diversity in the brain (Grant and Fransén, 2020) and spinal dorsal horn (Broadhead et al., 2020). We demonstrate here that a major locus of proteome complexity exists in the spinal lamina I nociceptive subfield and involves extensive postsynaptic GluN2 subtype diversity at the primary afferent-lamina I neuron synapse. Collectively, our data are consistent with a model wherein synapse-specific organization of distinct GluN2 subtypes serves as a fundamental subthreshold coding mechanism and a core determinant of how physiological sensory information is processed and transferred from the periphery to the brain. Future studies that scrutinize NMDAR subtype diversity and function at this principal sensory synapse in models of neuropathy and inflammation will undoubtedly advance our understanding of synaptic mechanisms underlying chronic pain associated with trauma or disease. Our findings also offer insight into complexity of glutamatergic synapses and their function in other regions of the CNS in health and disease.

Materials and methods
Animals Experiments were carried out using adult, male Sprague Dawley rats (from Charles River, P70-80, 325-350 g), and approved by the Hospital for Sick Children Animal Care Committee and in accordance with policies of the Canadian Council on Animal Care.
Data acquired from lamina I neurons included unitary excitatory postsynaptic excitatory postsynaptic currents (µEPSCs) or unitary excitatory postsynaptic potentials (µEPSPs) evoked by stimulating a single primary afferent synaptic connection to the lamina I neuron recorded. To isolate activation of an individual synaptic connection between a primary afferent axon and a lamina I neuron, the tip of an in-house designed and constructed (by GMP) fine bipolar stimulating electrode (consisting of two ∼50 µm diameter single nichrome wire electrodes (A-M Systems, Inc.) separated by ∼15 µm formvar insulation coating) was placed in the dorsal rootlet entry zone which contains axons of primary afferent neurons that are monosynaptically connected to lamina I neurons. Activation of a putative individual axon at the electrode tip consisted of single pulses of electrical stimulation (0.08 ms duration at 0.2 Hz delivered to the electrode via an S48 Grass Stimulator) with stimulus intensity increased gradually beginning at 0 stimulus intensity (minimal stimulation) until all-or-none unitary postsynaptic responses of the lamina I neuron recorded were detected at which point stimulus intensity (i.e., stimulus threshold) was no longer adjusted (see Results).
µEPSC recordings acquired from lamina I neurons were voltage-clamped at + 60 mv using pipettes containing Cs-gluconate ICS. In most of these voltage-clamp recordings µEPSCs were also acquired at a holding potential of −65 mv. µEPSPs were also acquired from lamina I neurons resting at −65 mv using pipettes containing K-gluconate ICS. Of note, prior to acquiring µEPSPs in a given recording we acquired first in the same recording µEPSCs in voltage-clamp mode (Vh: −65 mv). Rationale and further details of these experiments are provided in the Results.
Raw data were amplified using a MultiClamp 700A amplifier and a Digidata 1322A acquisition system sampled at 10 kHz, and analyzed with Clampfit 10.3 (Axon Instruments), Microsoft Office Excel, and Sigmaplot 11 software (Systat Software, Inc.). Junction potential was accounted for in all current-clamp experiments. Data from voltage-clamp experiments were discarded if the series resistance monitored in the recording changed by more than 15%.

Data analysis
Data from minimal stimulation experiments were considered bona fide minimal stimulation-elicited unitary postsynaptic responses and used for analysis only if the unitary responses in a given experiment exhibited a clear stimulus threshold (all-ornone postsynaptic events) determined upon increasing gradually stimulation intensity from 0 to the threshold stimulus intensity, a clear unimodal, normal distribution of the events, as well as a consistent latency with each stimulus (see Results). All recordings in every experiment were inspected visually to verify failure and non-failure responses. µEPSC kinetic parameters measured for analysis (including decay, amplitude, half-width, charge transfer, and rise time) were calculated from the mean µEPSC trace of all the non-failure responses in that experiment. The decay time constant(s) of the mean µEPSC (Vh: + 60 mv) was calculated in Clampfit by least-squares fitting with one or more exponentials. The best-fit of a given µEPSC decay was determined by comparing R 2 for the respective fit(s) of that decay. In current-clamp experiments, µEPSP kinetic parameters including peak amplitude, half-width, and area under the curve were also measured (see Results for details). Failure rate (%) of unitary events in a given minimal stimulation recording was calculated as the percentage of failures that occurred in that given recording.

Statistics
Comparisons between independent groups were carried out by two-tail t-test (no equal variance assumed). Comparisons between paired groups were carried out by paired t-test, comparisons on repeated measures were performed by repeated measures ANOVA. Significance was established at µ = 0.05, or at a corrected value following Bonferroni correction for multiple comparisons where appropriate. Kinetic data were normalized, scaled and analyzed in R by PCA and HC clustering, followed by gap-statistic to select the optimal number of clusters. HC was built using the euclidean distance method and the Ward linkage method.

Data availability statement
The original contributions presented in this study are included in this article/Supplementary material, further inquiries can be directed to the corresponding authors.

Ethics statement
The animal study was reviewed and approved by the Hospital for Sick Children Animal Care Committee.

Author contributions
GP conceived the study, designed the project, conducted the experiments, analyzed the data, and wrote the manuscript.
LG carried out statistical analysis and reviewed the manuscript. AM and MT advised on analysis design and reviewed the manuscript. MWS advised on analysis and project design and wrote the manuscript. All authors contributed to the article and approved the submitted version.  Figure 3A, represented by (i), is not significantly different from mean (± SEM) charge transfer of baseline µEPSCs held at + 60 mv from panel "(A)," represented by (iii). Mean (± SEM) charge transfer of µEPSCs held at −65 mv from Figure 3A, represented by (ii) and considered to be mediated predominantly by AMPAR-mediated charge transfer, is not significantly different from mean (± SEM) charge transfer of µEPSCs held at + 60 mv from panel "(B)" with TCN-201 present in aCSF, represented by (iv). (E) We reasoned that the difference in charge transfer between (i) and (ii) is the pure fast NMDAR component (white bar), and that the difference in charge transfer between (iii) and (iv) is the pure NMDAR component mediated by GluN2A (red bar). The lack of difference (p > 0.05) is therefore interpreted to indicate that the baseline NMDAR component of the fast µEPSC panels (A,B) is mediated exclusively by GluN2A. In panels (C-E), group data are mean ± SEM.

SUPPLEMENTARY FIGURE 9
Kinetics of µEPSCs recorded at the holding potential of −65 mv from Figure 3. (A) Late component of µEPSCs (Vh: −65 mv) exhibited a significant increase in half-width (p < 0.01, unpaired, two-tailed t-test) and charge transfer (p < 0.05, 0.01, 0.001, unpaired, two-tailed t-test) across the three groups. The dotted line, full line, the bottom and top of the box, and the whiskers of the plot represent mean, median, lower and upper quartiles, and 10th to 90th percentile, respectively. (B) Percent inhibition of the amplitude of the NMDAR component of µEPSCs held at −65 mv in fast, intermediate, and slow decay groups from Figure 3. The percent inhibition of the NMDAR component of µEPSCs from the "fast" group was similar to that of µEPSCs from the "intermediate" group, whereas the percent inhibition of the NMDAR component of µEPSCs from the "slow" group was less. Percent inhibition of the amplitude of the NMDAR component of µEPSCs held at −65 mv was calculated by dividing the difference between mean amplitude of the NMDAR component of µEPSCs held at −65 mv and the mean amplitude of the NMDAR component of µEPSCs held at + 65 mv by the mean amplitude of the NMDAR component of µEPSCs held at + 65 mv and multiplying the quotient by 100.

SUPPLEMENTARY FIGURE 14
Effects of D-APV or MK-801 on µEPSP bursts. (A) Mean (± SEM) µEPSP burst traces showing continuous recording during the 5 Hz train (gray arrows indicate stimulus artifacts throughout the train) during baseline (summated mean ± SEM trace), with D-APV (100 mM), followed by CNQX (10 mM). Given the continuous nature of these recordings, these traces include successes and failures. (B) Mean (± SEM) µEPSP burst traces showing continuous recording during the 5 Hz train (arrows indicate stimulus artifacts throughout the train) during baseline (summated mean ± SEM trace), with the NMDAR blocker, MK-801 (10 µM), followed by CNQX (10 mM). Given the continuous nature of these recordings, these traces include successes and failures.