Introduction

Ionotropic glutamate receptors (iGluRs), including the AMPA, NMDA, and kainate receptor subfamilies, are the primary mediators of fast excitatory neurotransmission in the vertebrate central nervous system1. Within this family, kainate receptors (KARs) occupy a unique functional niche. Rather than serving as the primary drivers of synaptic depolarization like AMPA receptors, KARs act as sophisticated modulators of neuronal circuits2,3,4. They are present at both the presynaptic and postsynaptic sites, where they regulate neurotransmitter release, influence neuronal excitability, and participate in various forms of synaptic plasticity5,6,7. The physiological importance of KARs is underscored by their involvement in higher brain functions such as network oscillations and memory formation, and their dysregulation is implicated in a range of pathologies, including epilepsy, major depressive disorder, and schizophrenia8,9,10,11.

KARs are encoded by five subunits, designated as GluK1 through GluK5, which are categorized as low and high-affinity receptors based on their glutamate binding affinity. Low-affinity KARs (GluK1-K3) can form homomeric or heteromeric assemblies, while high-affinity KARs (GluK4-5) associate with low-affinity subunits to form functional ion channels12,13. Among these, the GluK3 subunit is one of the least understood, yet it possesses distinct properties that suggest a highly specialized role. GluK3 is prominently expressed in specific brain regions, most notably the dentate gyrus and the CA3 region of the hippocampus, the key components of the mossy fiber pathway, as well as in the neocortex and cerebellum14,15,16. Recent behavioral analysis of knockout mice have specifically implicated GluK3-containing receptors in the expression of anxiety, highlighting their potential as therapeutic targets17. Functionally, GluK3-containing receptors are atypical; they exhibit low apparent sensitivity to glutamate despite the high affinity of their isolated ligand-binding domains (LBDs), a characteristic attributed to the extremely rapid desensitization of partially agonist-bound receptors16. This property suggests that GluK3 receptors are tuned to respond specifically to fast, high-concentration transients of glutamate16.

The functional properties of native KARs are shaped by a complex interplay with auxiliary proteins and endogenous ions. The discrepancy between the slow decay kinetics of native KAR-mediated synaptic currents and the rapid desensitization of recombinant receptors led to the discovery of the Neuropilin and Tolloid-like (Neto) proteins, Neto1 and Neto218. These single-pass transmembrane proteins are now established as essential KAR auxiliary subunits19,20,21,22,23. They profoundly modulate receptor gating by slowing desensitization and deactivation, altering recovery kinetics, and relieving voltage-dependent channel block by intracellular polyamines20,21,24,25,26. These effects are often dependent on both the specific KAR subunit and the Neto isoform involved, and Neto proteins also play a crucial role in receptor trafficking and synaptic targeting20. Intriguingly, Neto1 and Neto2 exhibit complementary expression patterns in the brain; Neto1 is highly expressed in the hippocampal CA3 region, whereas Neto2 is more abundant in the cerebellum and cortex, pointing toward distinct physiological roles21,23,27,28.

A second layer of modulation is provided by the divalent cation zinc (Zn2+). Zinc is an important neuromodulator that is packaged into synaptic vesicles by the transporter ZnT3 and co-released with glutamate at a subset of excitatory synapses, most prominently the hippocampal mossy fiber terminals29,30,31. Pioneering work has established that zinc uniquely potentiates GluK3-containing KARs, in stark contrast to its inhibitory effect on other KAR subtypes32. This potentiation arises from the binding of zinc to a specific, high-affinity site formed at the LBD dimer interface, which stabilizes the active state and markedly slows receptor desensitization32. The striking co-localization of GluK3, Neto1, and releasable zinc at the hippocampal mossy fiber-CA3 synapse suggests a functional convergence of these regulatory elements15,18,33,34. However, the physiological role of this system is complex and not fully resolved. Studies on GluK3 knockout mice have implicated these receptors in presynaptic forms of plasticity at mossy fiber synapses35. Paradoxically, the genetic removal of synaptic zinc via knockout of the ZnT3 transporter was found to have no effect on this form of presynaptic plasticity, creating an apparent contradiction36. This suggests that the interplay between these components is more nuanced than previously appreciated. To date, it remains unknown how these distinct modulatory layers, consisting of Neto proteins and zinc, interact. It is unknown if the Neto proteins alter the zinc sensitivity of GluK3 or if zinc influences the modulatory actions of the Neto proteins. Answering these questions is critical for building a foundational understanding of KAR signaling.

Towards this, we employed whole-cell patch-clamp electrophysiology to systematically characterize the functional interplay between GluK3, Neto1, Neto2, and zinc in a heterologous expression system. A central component of this strategy is the use of a GluK3 (D759G) mutant. Based on the established role of Asp759 in coordinating zinc, this mutation specifically ablates the zinc-binding site32. This mutant serves as a tool to mechanistically dissect the effects of zinc from the broader modulatory actions of the Neto proteins. Furthermore, to provide a structural correlate for its functional properties, we used cryo-electron microscopy (cryo-EM) to visualize the antagonist-bound (D759G) mutant receptor. This combined functional and structural approach reveals a sophisticated, tripartite regulatory mechanism that dictates GluK3 receptor function and provides a detailed biophysical framework for interpreting its complex roles in the brain.

Results

Neto1 and Neto2 differentially regulate the gating of wild-type GluK3 receptors

To define the baseline modulatory effects of Neto proteins on GluK3, we first performed whole-cell patch-clamp recordings from HEK293 cells expressing wild-type (WT) GluK3 receptors alone or co-expressed with either Neto1 or Neto2. Application of a saturating concentration of glutamate (30 mM) to cells expressing GluK3 alone elicited rapidly activating and desensitizing currents, with a mean weighted desensitization time constant (τdes) of 4.80 ± 0.23 ms (n = 8) (Fig. 1A, B and Table 1). Co-expression with Neto1 significantly slowed this desensitization to 6.24 ± 0.47 ms (n = 11; *p = 0.03) (Fig. 1A, B and Table 1), while co-expression with Neto2 had an even more pronounced effect, increasing τdes to 11.59 ± 0.53 ms (n = 9; ****p < 0.0001) (Fig. 1A, B and Table 1). This demonstrates that both auxiliary subunits stabilize the active state of the receptor, with Neto2 having a more potent effect on slowing desensitization.

Fig. 1: Neto-mediated modulation of GluK3 receptor desensitization kinetics and rectification.
Fig. 1: Neto-mediated modulation of GluK3 receptor desensitization kinetics and rectification.The alternative text for this image may have been generated using AI.
Full size image

A Representative whole-cell current traces evoked by 30 mM glutamate (100 ms application) from HEK293 cells expressing GluK3 alone or co-expressed with Neto1 or Neto2. The inset shows a magnified view of the traces to facilitate visualization of differences in current kinetics (all normalized traces are shown using the same time scale). B Comparison of mean weighted desensitization time constants (τ des, ms) illustrated for whole-cell current traces (GluK3 n = 8, GluK3+Neto n = 11, GluK3+Neto2 n = 9) shown in (A). C Mean I-V relationships for GluK3 (n = 4) expressed alone or with Neto1 (n = 4) or Neto2 (n = 5) in the presence of 30 mM glutamate. GluK3 alone exhibits negligible outward currents, whereas co-expression with Neto1 or Neto2 produces prominent outward currents (inset, all normalized traces are shown using the same time scale). Whole-cell currents were normalized to the peak amplitude at –90 mV to allow comparison of rectification differences between depolarized and hyperpolarized potentials. D Pooled data of rectification index (RI, +90 mV/−90 mV) calculated from the corresponding I–V curves. Data are presented as mean ± s.e.m. Statistical significance was assessed using an unpaired two-tailed t-test (*p < 0.05;**p < 0.01; ***p < 0.001;****p < 0.0001).

Table 1 Desensitization kinetics and biophysical properties of GluK3 and GluK3 (D759G) EM in the presence or absence of auxiliary proteins Neto1 or Neto2
Full size table

GluK3 receptors are known to exhibit strong inward rectification due to a voltage-dependent block by intracellular polyamines such as spermine. We quantified this by measuring the rectification index (RI), which is the ratio of the outward current at +90 mV to the inward current at −90 mV. As expected, GluK3 WT receptors were strongly rectifying, with a very low RI of 0.04 ± 0.02 (n = 4) (Fig. 1C, D and Table 1). Co-expression with either Neto1 or Neto2 significantly relieved this block, increasing the RI to 0.25 ± 0.04 (n = 4; *p = 0.003) (Fig. 1C, D and Table 1) and 0.19 ± 0.03 (n = 5; *p = 0.006) (Fig. 1C, D and Table 1), respectively. This suggests that Neto proteins alter the conformation of the ion channel pore or the access pathway for polyamines, a common feature of kainate receptor auxiliary subunit modulation37.

While both Neto proteins slowed desensitization and relieved channel block, they exhibited strikingly divergent effects on receptor recovery from desensitization, a key parameter governing the ability of a receptor to respond to high-frequency stimulation. Using a paired-pulse protocol, we found that Neto1 significantly accelerated the recovery from desensitization (τrecovery = 0.37 ± 0.11 s, n = 5, **p = 0.002) (Fig. 2A–C and Table 1) compared to GluK3 WT alone (τrecovery = 0.72 ± 0.27 s, n = 3) (Fig. 2A–C and Table 1). In stark contrast, Neto2 profoundly decelerated the recovery process, extending the time constant to 2.31 ± 1.3 s, n = 6, **p = 0.009 (Fig. 2A–C and Table 1). This opposing regulation of recovery kinetics represents a critical functional divergence between the two Neto isoforms, implying that they stabilize distinct conformational states of the receptor.

Fig. 2: Modulation of GluK3 recovery from desensitization by Neto proteins.
Fig. 2: Modulation of GluK3 recovery from desensitization by Neto proteins.The alternative text for this image may have been generated using AI.
Full size image

A Representative whole-cell current traces from HEK293 cells expressing GluK3 alone (n = 4) or co-expressed with Neto1 (n = 5) or Neto2 (n = 6), illustrating recovery from desensitization. Recovery was assessed using paired applications of 30 mM glutamate with variable inter-pulse intervals (50 ms–2 s; longer intervals were also tested). Peak currents evoked by the second pulse were normalized to the corresponding first response. All normalized traces are shown using the same time scale. B Time course of recovery from desensitization plotted as percentage recovery versus inter-pulse interval (log₁₀ scale) for the indicated conditions. Recovery curves were fitted using a one-phase association nonlinear regression model. Data represent mean ± s.e.m. from 3 to 6 independent recordings. C Summary of recovery time constants (τ_recovery) derived from one-phase association nonlinear regression fits to the mean recovery curves shown in (B), with values reported as mean ± s.e.m. across independent recordings, revealing Neto-dependent modulation of recovery kinetics. Data are shown as mean ± s.e.m. Statistical significance was determined using an unpaired two-tailed t-test (*p < 0.05;**p < 0.01;***p < 0.001;****p < 0.0001).

Interestingly, despite these profound effects on gating, neither Neto1 nor Neto2 significantly altered the activation kinetics of the WT receptor. The 10–90% rise time of glutamate-evoked currents remained unchanged for GluK3 WT in the presence of Neto1 (1.91 ± 0.22 ms, n = 11; p = 0.87) or Neto2 (3.49 ± 0.49 ms, n = 8; p = 0.19) in comparison to GluK3 WT alone (2.33 ± 0.3 ms, n = 3) (Table 1). A similar lack of Neto-dependent modulation of activation kinetics was also observed for the D759G mutant (see below), despite pronounced effects on desensitization. Co-expression of the mutant with Neto1 (2.67 ± 0.18 ms) or Neto2 (2.84 ± 0.14 ms) resulted in rise times that were comparable to, or slightly faster than, the mutant receptor alone (3.12 ± 0.23 ms). This dissociation between the regulation of activation (rise time) and inactivation (desensitization) suggests that Neto proteins primarily stabilize the open state rather than the initial opening transition. To confirm that the observed functional effects were due to a direct physical interaction, we performed reciprocal pull-down experiments using His-tagged GluK3 and Strep-tagged Neto1 and Neto2 proteins. In both configurations, Neto1 and Neto2 robustly co-purified with GluK3, confirming the formation of a stable and specific complex between GluK3 and Neto proteins in our expression system (Supplementary Fig. 1A, B).

Zinc potentiation of GluK3 is uniquely modulated by Neto auxiliary subunits

Having established the baseline effects of Neto proteins, we next investigated how they interact with the endogenous zinc cation to modulate the properties of the GluK3 receptor. We first confirmed the known potentiation of WT GluK3 by zinc32. For GluK3 WT receptors expressed alone, zinc produced a 2.02 ± 0.32-fold slowing of the desensitization rate (n = 5) (Fig. 3A, D and Table 2) and a modest 1.7 ± 0.17-fold increase in peak current amplitude (n = 5) (Fig. 3E, H and Table 2). In the presence of Neto1, the effect of zinc on desensitization was significantly attenuated; zinc produced only a 1.34 ± 0.09-fold further slowing of τdes (n = 6, **p < 0.01) (Fig. 3B, D and Table 2) and had no significant effect on peak current amplitude (1.03 ± 0.05-fold change, n = 6) (Fig. 3F, H and Table 2). As mentioned above, the interaction between Neto2 and zinc was markedly different and synergistic. While Neto2 co-expression alone did not cause a significant increase in peak current amplitudes, the co-application of Neto2 and zinc resulted in a powerful, 3.2 ± 0.61-fold facilitation of the peak current (n = 4, ****p < 0.0001) (Fig. 3G, H and Table 2). This effect showed a strong trend toward significance in the raw peak amplitudes (*p = 0.052) and was highly significant upon normalization. This facilitation was significantly greater than that seen with zinc alone or with Neto1 and zinc. Zinc also continued to slow the desensitization of GluK3/Neto2 complexes, producing a further 1.57 ± 0.1-fold increase in τdes (n = 5, **p < 0.01) (Fig. 3C, D and Table 2). This suggests a cooperative allosteric interaction, where the binding of both Neto2 and zinc induces a receptor conformation with higher efficacy than can be achieved by either modulator alone.

Fig. 3: Zinc potentiation of GluK3 desensitization is differentially modulated by Neto subunits.
Fig. 3: Zinc potentiation of GluK3 desensitization is differentially modulated by Neto subunits.The alternative text for this image may have been generated using AI.
Full size image

A–C Mean weighted τdes of GluK3 alone (n = 5) and modulated by Neto1 (n = 6) or Neto2 (n = 5) currents were shown for 30 mM glutamate application and 100 μM Zn, along with 30 mM glutamate. Insets show representative normalized current traces for each condition. All normalized traces are shown using the same time scale. D Normalized histogram presents the effect of 100 μM zinc on GluK3 receptors in the presence of Neto1 or Neto2. E–G Peak glutamate-evoked current amplitudes recorded from cells expressing GluK3 alone (n = 5) or with Neto1 (n = 6) or Neto2 (n = 6). In each panel (above), the inset shows normalized currents as a visual representation (all normalized traces are shown using the same time scale). Neto2 had a significant facilitatory effect on GluK3-mediated currents. H Normalized histogram of peak current amplitudes from GluK3 receptors expressed in the absence or presence of Neto proteins. Data are shown as mean ± s.e.m. Statistical significance was assessed using a paired two-tailed t-test for comparisons in (A, B, E–G), and an unpaired two-tailed t-test for comparisons of normalized mean weighted desensitization time constants (τdes). Significance is indicated as *p < 0.05; **p < 0.01;***p < 0.001;****p < 0.0001; ns not significant.

Table 2 Effect of Zn2+ potentiation on desensitization and functional properties of GluK3 and GluK3(D756G) EM receptors, with or without Neto1 or Neto2
Full size table

The D759G mutation uncouples zinc sensitivity from Neto modulation

We noted that standard extracellular solutions often contain trace amounts of contaminating zinc (nanomolar range), which can tonically inhibit other glutamate receptors such as NMDARs. However, the potentiation of GluK3 requires micromolar zinc concentrations. Therefore, we define our “control” condition as the absence of experimentally added zinc. To rigorously isolate the specific potentiation effect, we utilized the D759G mutant. This mutation replaces the key zinc-coordinating aspartate residue at the LBD dimer interface with a neutral glycine residue32,38. As predicted, this single mutation had profound effects on receptor function. The (D759G) mutant exhibited a baseline desensitization rate that was more than three times slower (τdes = 15.23 ± 0.92 ms, n = 11) than WT GluK3 (τdes = 4.8 ± 0.23 ms, n = 8; ****p < 0.0001) (Supplementary Fig. 1C and Table 1). This is consistent with the removal of electrostatic repulsion between Asp759 and Asp730 at the dimer interface, resulting in a more stable LBD assembly32.

Crucially, while the (D759G) mutant ablated the potentiating effect of zinc, it unmasked a secondary, inhibitory action. In stark contrast to its effect on WT receptors, co-application of 100 µM zinc with glutamate to GluK3 (D759G) receptors caused a significant acceleration of desensitization (Fig. 6A and Table 2) and a reduction in peak current amplitude (Fig. 6E and Table 2). This result provides strong evidence that Asp759 is essential for the potentiating effect of zinc and reveals a previously uncharacterized inhibitory zinc-binding site on the GluK3 receptor.

Despite the profound changes to the LBD interface and zinc sensitivity, the (D759G) mutant was still robustly modulated by Neto proteins. Co-expression with Neto1 and Neto2 further slowed the already slow desensitization of the D759G mutant to 38.18 ± 2.67 ms with Neto1, n = 6, ***p = 0.0001; and to 51.24 ± 7.03 ms with Neto2, n = 10, ****p < 0.0001 (Fig. 4A, B and Table 1). Neto proteins also relieved GluK3 (D759G) mutant receptor’s inward rectification (RI with Neto1: 0.27 ± 0.02, n = 5, ****p < 0.0001; RI with Neto2: 0.14 ± 0.035, n = 8, *p = 0.0094) (Fig. 4C, D and Table1), mirroring their effects on the WT receptor (Fig. 1C, D and Table 1). Most importantly, the signature divergent effects of Neto proteins on the recovery kinetics were fully preserved and Neto1 accelerated recovery of the (D759G) mutant (τrecovery = 0.23 ± 0.05 s, n = 6; ****p < 0.0001) (Fig. 5A–C and Table 1), while Neto2 decelerated it (τrecovery = 1.72 ± 0.40 s, n = 6; ***p < 0.001) (Fig. 5A–C and Table 1). This demonstrates that the core modulatory machinery of Neto proteins is independent of the Asp759 residue and the zinc-binding pocket. Furthermore, this inhibitory effect of zinc was itself subject to modulation by Neto proteins. In the presence of Neto1, the zinc-induced acceleration of desensitization was diminished, whereas in the presence of Neto2, it was retained (Fig. 6B–D and Table 2).

Fig. 4: Neto-mediated modulation of the GluK3(D759G) receptor desensitization and rectification.
Fig. 4: Neto-mediated modulation of the GluK3(D759G) receptor desensitization and rectification.The alternative text for this image may have been generated using AI.
Full size image

A Representative whole-cell current traces evoked by 30 mM glutamate (100 ms application) from HEK293 cells expressing GluK3(D759G) alone or in combination with Neto1 or Neto2 (all normalized traces are shown using the same time scale). B Summary of mean weighted desensitization time constants (τdes) (GluK3 D759G n = 11, GluK3 D759G+Neto1 n = 6, GluK3 D759G +Neto2 n = 10) derived from the current traces shown in (A). C Average current–voltage (I–V) relationships for GluK3(D759G) expressed alone or co-expressed with Neto1 or Neto2 in the presence of 30 mM glutamate. GluK3(D759G) alone (n = 7) shows minimal outward currents, whereas co-expression with Neto1 (n = 5) or Neto2 (n = 8) markedly enhances outward currents (inset, all normalized traces are shown using the same time scale). Whole-cell currents were normalized to the peak amplitude at –90 mV to enable comparison of rectification properties between depolarized and hyperpolarized voltages. D Quantification of rectification index (RI, +90 mV/–90 mV) obtained from the corresponding I–V relationships. Error bars indicate mean ± s.e.m. Statistical significance was assessed using an unpaired two-tailed t-test (****p < 0.0001; ***p < 0.001).

Fig. 5: Modulation of GluK3 (D759G) receptor recovery from desensitization by Neto proteins.
Fig. 5: Modulation of GluK3 (D759G) receptor recovery from desensitization by Neto proteins.The alternative text for this image may have been generated using AI.
Full size image

A Representative whole-cell current traces from HEK293 cells expressing GluK3(D759G) alone (n = 6) or co-expressed with Neto1 (n = 6) or Neto2 (n = 6), illustrating recovery from desensitization. Recovery was assessed using paired applications of 30 mM glutamate with variable inter-pulse intervals (50 ms–2 s; longer intervals were also tested). Peak currents evoked by the second pulse were normalized to the corresponding first response. All normalized traces are shown using the same time scale. B Time course of recovery from desensitization plotted as percentage recovery versus inter-pulse interval (log₁₀ scale) for the indicated conditions. Recovery curves were fitted using a one-phase association nonlinear regression model. Data represent mean ± s.e.m. from 3 to 6 independent recordings. C Summary of recovery time constants (τRecovery) obtained from the one-phase association fits shown in (B), with values reported as mean ± s.e.m. across independent recordings (n = 5–8), revealing Neto-dependent modulation of recovery kinetics. Data are shown as mean ± s.e.m. Statistical significance was determined using an unpaired two-tailed t-test (***p < 0.001;****p < 0.0001).

Fig. 6: Zinc potentiation of GluK3(D759G) desensitization is differentially modulated by Neto subunits.
Fig. 6: Zinc potentiation of GluK3(D759G) desensitization is differentially modulated by Neto subunits.The alternative text for this image may have been generated using AI.
Full size image

A–C Mean weighted τdes of GluK3(D759G) alone (n = 9) and modulated by Neto1 (n = 5) or Neto2 (n = 6) currents were shown for 30 mM glutamate application and 100 μM Zn, along with 30 mM glutamate. In each panel, an inset shows normalized currents as a representation. All normalized traces are shown using the same time scale. D. Normalized histogram presents the effect of 100 μM zinc on the desensitization kinetics of GluK3(D759G) receptors in the presence of Neto1 or Neto2. E–G Peak glutamate-evoked current amplitudes recorded from cells expressing GluK3 alone (n = 9) or with Neto1 (n = 5) or Neto2 (n = 8). In each panel (above), the inset shows normalized currents as a visual representation. Neto2 had a significant facilitatory effect on GluK3(D759G)-mediated currents. H Histogram displays the normalized peak amplitudes from GluK3 (D759G) receptors with and without Neto proteins. Data are shown as mean ± s.e.m. Statistical significance was assessed using a paired two-tailed Student’s t test for comparisons in (A, B, E–G), and an unpaired two-tailed Student’s t-test for comparisons of normalized mean weighted desensitization time constants (τdes). Significance is indicated as *p < 0.05; **p < 0.01;***p < 0.001;****p < 0.0001; ns not significant.

This uncoupling also revealed a critical mechanistic insight related to receptor activation. While Neto proteins had no effect on the rise time of the “labile” WT receptor, the rise time of the (D759G) mutant (3.12 ± 0.23 ms, n = 11) was reduced by co-expression with Neto1 (2.67 ± 0.18 ms, n = 6; p = 0.20) and Neto2 (2.84 ± 0.14, n = 10; p = 0.31) (Table 1). This suggests that the intrinsic stability of the LBD dimer is a key determinant of the functional outcome of Neto modulation.

The D759G mutation stabilizes the LBD layer

Our cryo-EM analysis provides a structural rationale for these functional observations. We purified GluK3 (D759G) mutant receptors and determined their structure by single-particle cryo-EM. Structures were solved in the presence of the competitive antagonist UBP310 and the channel blocker spermine. The construct used for expression was based on a previously developed stabilized version of GluK338. For the GluK3 (D759G) receptor in the presence of UBP310, the final 3D reconstruction was resolved to a global resolution of ~5.9 Å (Supplementary Fig. 2). While this resolution precludes detailed analysis of side-chain interactions, the overall map revealed the canonical three-layered architecture of iGluRs (Fig. 7B–D). Importantly, comparison of our GluK3 (D759G) structure with previously published structures of antagonist-bound WT GluK3 (PDB IDs: 6JFZ, 6L6F) showed that the (D759G) mutation induces a more compact LBD arrangement38. Comparison of our GluK3 D759G structure with previously published WT GluK3 structures reveals that the mutation induces a 3–6 Å compaction of the LBD dimer interface (Fig. 7G–L). This structural tightening likely mimics the “clamped” state induced by zinc binding in the WT receptor. By physically constraining the LBD dimer in a more compact arrangement, the D759G mutation, and by extension, zinc binding, likely raises the energetic barrier for the conformational decoupling that leads to desensitization. Superimposition of the LBD tetramers revealed subtle but significant rotations of the subunits in the mutant structure. This observation provides direct structural evidence for the “tighter packing” and enhanced stability of the LBD dimer. This also explains why the mutant exhibits such profoundly slowed desensitization kinetics even in the absence of zinc.

Fig. 7: Overview of the GluK3 D759G construct, cryo-EM map, and structural rearrangements in the ligand-binding domain relative to the wild-type receptor.
Fig. 7: Overview of the GluK3 D759G construct, cryo-EM map, and structural rearrangements in the ligand-binding domain relative to the wild-type receptor.The alternative text for this image may have been generated using AI.
Full size image

A Schematic representation of GluK3 mutant (D759G) construct. B, C CryoEM density map segmented and colored as different chains (labeled as A, B, C, D) at 5.9 Å for GluK3 bound to Spermine and antagonist UBP310. Front and side views for the color-segmented map. D Fitted atomic model colored to represent four receptor subunits. E, F Side views of the ligand binding domain dimer interface of GluK3 WT and GluK3 (D759G) showing location of mutant glycine at 759th position indicated by red sphere. G, J Superimposition of LBD tetramers from GluK3D759GUBP310+spermine (green), GluK3WTUBP310 (red), GluK3WTUBP301 (yellow), depicting the degree and direction of LBD tetramer movement based on the angles calculated between Cα atoms of leu405 (shown as spheres) located at the top of the LBD. H, I, K, L Superimposition of LBD dimers BC and AD from GluK3D759GUBP310+spermine (green), GluK3WTUBP310 (red), GluK3WTUBP301 (yellow). Distances between the Cα atoms of S742 and I639 are measured to represent the S1 and S2 domain movement of all LBD dimers.

A critical finding was the conformational heterogeneity within the LBD layer. Similar to antagonist-bound WT structures38,39, the LBDs in our D759G map did not adopt a uniform, classical C2-symmetric closed state. Instead, while two subunits formed a canonical “back-to-back” dimer, the other two remained in a “desensitized-like” conformation, with their LBDs swung apart. This persistence of a mixed conformational state, even in a receptor stabilized by both a mutation and an antagonist, powerfully illustrates the intrinsic lability of the GluK3 LBD assembly. This low-resolution structural data support the functional observation that the D759G mutation stabilizes the LBD dimer interface, while also highlighting the inherent conformational flexibility of the GluK3 receptor.

Discussion

Our study unveils a sophisticated, tripartite regulatory system that governs the function of the GluK3 kainate receptor. Our findings demonstrate that GluK3 is not a static entity but a dynamic signaling molecule that integrates information from its primary agonist, glutamate, with modulatory inputs from Neto auxiliary subunits and the endogenous zinc ion. While the physiological relevance of this system requires further in vivo investigation, its components, GluK3, Neto proteins, and synaptically released zinc are all known to be present at key central synapses, such as the hippocampal mossy fiber-CA3 synapse, a locus for synaptic plasticity and memory formation15,33,34,18,35,. Our work provides a foundational biophysical framework for understanding how the functional diversity of GluK3 receptors is generated at such synapses.

A central finding of our work is that Neto1 and Neto2, despite their structural similarity, differentially regulate the gating of GluK3 receptors. While both slow desensitization and relieve polyamine block, their opposing actions on recovery from desensitization, acceleration by Neto1 and deceleration by Neto2, imply distinct underlying mechanisms. This divergence suggests that neurons can likely tune the temporal filtering properties of KAR-mediated currents by altering the relative expression of Neto isoforms. A synapse enriched in Neto1-associated KARs would be poised to faithfully follow high-frequency inputs. In contrast, a synapse dominated by Neto2-associated KARs would integrate signals over longer timescales, profoundly impacting short-term plasticity40.

Recent cryo-EM structures of the GluK2-Neto2 complex show that Neto2 makes extensive contacts with both the ATD and LBD layers of the receptor37,41. We speculate that the differential effects of Neto1 and Neto2 arise from subtle differences in these multi-domain interactions. While both isoforms likely stabilize the LBD dimer in the active state (slowing desensitization), their distinct contacts with the ATD and the linkers between domains could create unique energy landscapes for the conformational transitions of the KARs. The faster recovery induced by Neto1 might suggest it preferentially stabilizes an intermediate state that facilitates a rapid return to the resting state. Conversely, the slower recovery with Neto2 suggests it may more strongly stabilize the fully desensitized conformation itself, likely creating a deeper energetic well from which the receptor must escape. Our use of the (D759G) mutant provided a powerful tool to dissect the layers of modulation. The co-operation observed between Neto2 and zinc is a distinct property that could not be predicted from their individual actions. We propose a cooperative allosteric model where two distinct modulatory sites communicate to produce an emergent functional outcome. Zinc, by binding at the base of the LBD dimer, “clamps” the D1 lobes together32, while Neto2, through its extensive contacts on the outer surface of the ATD and LBD layers37, provides an external scaffold. Together, these two modulators appear to lock the receptor in a high-efficacy conformation that is more readily activated, leading to the observed facilitation of peak current. This mechanism could allow synapses to strongly amplify KAR signaling, specifically when high levels of glutamate and zinc are co-released during periods of intense activity.

Perhaps the most surprising finding of this study is that the (D759G) mutation does not simply render GluK3 insensitive to zinc, but rather unmasks a secondary, inhibitory modulatory site. This suggests that GluK3 possesses at least two functionally distinct zinc-binding sites: (1) the high-affinity potentiating site at the LBD dimer interface involving D759, and (2) a lower-affinity, inhibitory site whose action is normally overshadowed by the dominant potentiation. The differential modulation of this inhibitory effect by Neto1 (which blunts it) and Neto2 (which does not) adds yet another layer of complexity. The molecular identity of this secondary site remains to be determined, but conserved histidine and acidic residues in the extracellular domains (e.g., H80, H93, H243, E96, or E297) represent plausible candidates that need future investigation. The conformational heterogeneity observed even in our antagonist-bound mutant structure, which proved challenging for high-resolution reconstruction, underscores the profound intrinsic lability of the GluK3 LBD assembly. This inherent instability provides a structural rationale for why GluK3 is so heavily dependent on regulation by auxiliary subunits and ions to achieve stable gating. Recent high-resolution structural work on other KARs has revealed a complex desensitization pathway involving large-scale lateral rotation of the LBDs, leading to distinct “shallow” and “deep” desensitized states that differ from those of AMPA receptors42,43. Our data are consistent with this model, suggesting that the desensitized state for KARs, and perhaps for GluK3 in particular, represents an exceptionally deep and accessible energy well. We propose that this intrinsic instability creates a highly responsive platform that is sensitive to regulation. This inherent propensity to desensitize necessitates the evolution of a rich repertoire of modulators, including auxiliary subunits like Neto proteins and endogenous ions like zinc, to control receptor function and expand its signaling capacity.

Our findings also have significant implications for understanding both normal brain function and disease. The differential expression of Neto1 and Neto2 across brain regions suggests that the rules of KAR modulation are not uniform throughout the CNS18. At the mossy fiber-CA3 synapse, the high expression of Neto1 and the abundance of synaptic zinc create a unique regulatory environment. The apparent contradiction between the potent in vitro modulation of GluK3 by zinc and the lack of effect of zinc removal on presynaptic plasticity in vivo36 may be partially explained by our findings. For instance, the presence of Neto1, which we show attenuates zinc potentiation, could render presynaptic GluK3 receptors less sensitive to ambient synaptic zinc under certain conditions. This complex regulatory axis also represents a novel, unexplored nexus in neurological and psychiatric disorders. In temporal lobe epilepsy, where mossy fiber sprouting and aberrant zinc release are common, the synergistic potentiation of GluK3 by Neto2 and zinc may contribute to network hyperexcitability44. In conditions like depression and schizophrenia, where alterations in both zinc homeostasis45 and KAR gene expression46 have been reported, the interplay we describe here could be a key factor in synaptic dysregulation. Targeting the specific interactions between KARs, Neto proteins, and their ionic environment may offer a more nuanced therapeutic strategy than direct receptor agonism or antagonism.

In conclusion, our work examines the complex regulation of the GluK3 kainate receptor, revealing a hierarchical system of control that involves intrinsic subunit properties, differential gating by Neto auxiliary subunits, and allosteric tuning by synaptic zinc. Our work presents a distinct framework for understanding how the functional diversity of GluK3 receptors is generated at central synapses, offering potential avenues for the development of therapies for disorders of glutamatergic signaling.

Materials and methods

Electrophysiology

The DNA constructs for the rat GluK3 (GenBank: NP_058721.1) and the rat GluK3 (D759G) are Q version variants. The rat Neto1 (GenBank: NP_001100318.1) and rat Neto2 (GenBank: NP_001099859.1) constructs were used. All constructs were cloned into the pRK5 (rat GluK3 WT) or pEGBacMam vector (rat GluK3 (D759G), rat Neto1, rat Neto2). A fluorescent tag was added to the C-terminal end of each construct: EGFP was used for GluK3 and GluK3 D759G EM (with K2 signal sequence at the N-terminus), while mRuby was used for the rat Neto1 and rat Neto2 constructs. HEK293 cells were cultured on uncoated cover slips placed in a 35 mm polystyrene dish at 37 °C under 5% CO₂ in DMEM (high glucose; catalog no. 1210006) supplemented with NaHCO3, 10% fetal bovine serum (FBS; Gibco, catalog no. 16000044), and 1% penicillin–streptomycin (Gibco, catalog no. 15140122). The cells were co-transfected with wild-type or mutant GluK3 receptors using the calcium phosphate transfection kit (CalPhosTM, catalog no. 631312). Cells showing moderate EGFP/mRuby fluorescence and 5–6 pF capacitance were selected to record currents. The cells were voltage-clamped at −80 mV, and the ligand was applied using an ultrafast perfusion system comprised of a two-barrel theta pipette mounted on a piezoelectric device. The entire setup was controlled by PatchMaster software, which filtered and digitized the responses at 3 kHz and 20 kHz, respectively.

The borosilicate glass capillaries (Sutter, O.D.: 1.0 mm, I.D.: 0.050 mm, and 10 cm length, ITEM# BF150-86-10) were pulled into recording pipettes using a micropipette puller (Sutter, Model O-1000) and fire-polished with a microforge developed in-house. The polished pipettes, having a resistance of 2–3 MegaΩ, were filled with an intracellular solution (100 mM CsF, 30 mM CsCl, 10 mM HEPES, 5 mM EGTA, 4 mM NaCl, 2 mM Na2ATP, pH 7.2, and osmolarity of 290–300 mOsm/L), and used for recordings. Continuous perfusion of an extracellular solution (150 mM NaCl, 10 mM HEPES, 2.8 mM KCl, 2 mM CaCl2, 1 mM MgCl2, 10 mM glucose, pH 7.3, and osmolarity of 295–300 mOsm/L) was performed at a flow rate of 0.2 ml/min, accompanied by recording. After attaining the voltage clamp, the cell was detached from the cover slip and placed opposite to a two-barrel theta pipette perfusing ligand and control solution in two separate streams. 30 mM glutamate was applied for 100 ms, 500 ms, and 1000 ms, as required by the experiment, and currents were recorded 60 s after the whole-cell configuration was established. During Zn potentiation experiments, a Zn-containing bath solution was pre-applied to measure the potentiation currents in the presence of 100 µM Zn and 30 mM glutamate.

Data analysis

The raw files (dat) were exported as ITX files and converted into abf files using the ABF utility to render them compatible with Clampfit 11.2 (Molecular Devices). Exponential fitting was performed to determine the rate of desensitization (τ des) using values corresponding to 90% of the peak amplitude in pClamp. The mean weighted τ des was obtained by performing a bi-exponential fit (Chebyshev algorithm, built-in Clampfit 11.2, Molecular Device), which was used to calculate mean τ desensitization by using:

(A1* τ 1 + A2* τ 2)/(A1 + A2)

A1 = Amplitude (initial magnitude) of the first exponential component

τ1 = Time constant of the first component

A2 = Amplitude of the second exponential component

τ2 = Time constant of the second component

To study recovery from desensitization, a paired-pulse protocol was employed in which the amplitude of the test pulse was normalized to that of the preceding desensitizing pulse and plotted against the time interval between the paired pulses. The time course of recovery from the desensitized state was fitted using a one-phase association exponential function (built-in function in GraphPad Prism 8.4). Recovery time constants (τRecovery) were obtained from these fits. Log-transformed recovery times were then plotted against the corresponding percentage recovery values, and the mean τRecovery values were used to generate bar graphs in Prism.

Construct design, expression and purification of GluK3 D759G receptors for structural studies

The native signal peptide of GluK3 was substituted with that of GluK2 and sub-cloned into the pEGBacMam vector for baculovirus-based expression in mammalian cells in order to improve surface expression. A thrombin recognition site and linker sequence (GLVPRGSAAAA) were inserted between GluK3 and the coding sequence for the A207K non-dimerizing EGFP mutant, with a C-terminal octa-histidine (His8) affinity tag, in order to screen constructs via fluorescence detection and for affinity purification. The GluK3 construct is truncated at C-ter ∆826 to increase its solubility and stability. Additionally, cysteine residues at positions 86, 305 are mutated to threonine, while cysteine 547 is changed to valine38. In order to improve the LBD dimer interface stability, an aspartate residue (D759) present at the LBD dimer interface is mutated to glycine to neutralize the electrostatic repulsion between Asp759 and Asp730 of each GluK3 subunit which are in close proximity to Asp730 and Asp759 from the partner protomer.

HEK293 GnTIˉ (ATCC) suspension culture at a density of 3×106 cells/ml was infected with P3 baculovirus expressing GluK3 D759G with multiplicity of infection (MOI) value of 2. Sodium butyrate was added to a final concentration of 10 mM, 20 h post-infection, and the culture was shifted to 30 °C for protein expression. The culture was harvested 70 h post-infection, and the cell pellet was stored at −80 °C till further use. The cell pellet was resuspended in chilled buffer (20 mM Tris pH 8.0, 150 mM NaCl) containing protease inhibitor cocktail and subjected to ultrasonication with a medium probe maintaining a temperature range of 6–12 °C (QSonica sonicator, 3 cycles of 90 s at power 7; 10-s pulse on, 10-s pulse off). Low-speed centrifugation was employed to clarify the lysate, and the supernatant was used to obtain membranes by ultracentrifugation (39,000 rpm for 60 min).

The membrane pellet was resuspended in buffer (20 mM Tris pH 8.0, 150 mM NaCl) and homogenized with the help of a Dounce homogenizer, followed by the addition of 30 mM n-dodecyl-β-D-maltopyranoside (DDM) and 6 mM cholesteryl hemisuccinate (CHS) for solubilization at 4 °C. The solubilized homogenate was subjected to centrifugation at 12,000 rpm at 4 °C for 45 min to remove the insoluble fraction. Cobalt-charged Talon resin (~3 mL) and imidazole (final concentration, 10 mM) were added to the supernatant and incubated for batch binding on a magnetic stirrer for 3 h at 4 °C. Beads were collected in a column and washed with ice-cold buffer (20 mM Tris, 150 mM NaCl, 0.75 mM DDM, and 0.03 mM CHS, pH 8) using an imidazole step gradient corresponding to 10 mM and 40 mM in successive steps until zero absorbance at 280 nm was achieved. The bound protein was eluted with elution buffer (20 mM Tris, 150 mM NaCl, 250 mM imidazole, 0.75 mM DDM, and 0.03 mM CHS, pH 8.0).

The eluates containing protein were pooled and concentrated to 7 mg/mL, followed by size exclusion chromatography (Superose 6 10/300 GL) using 20 mM Tris, 150 mM NaCl, 0.75 mM DDM, and 0.03 mM CHS at pH 8. Fractions containing purified protein were pooled and concentrated to 2.5 mg/ml to limit the concentration of DDM up to 3.5 mM (Supplementary Fig. 1D, E). Small volume aliquots of the protein were flash-frozen in liquid nitrogen and stored at −80 °C till further use.

GluK3 and Neto pull-down assay

HEK GnTI- cells were co-transfected with His-tagged GluK3 and StrepII-tagged Neto constructs. Cells were harvested and lysed in a buffer containing 20 mM Tris pH 8, 150 mM NaCl, 1% glycerol, protease inhibitor complex (Sigma), and 30 mM DDM. The supernatant was incubated with either Co-Talon (Takara) or StrepTactinXT beads (IBA). After extensive washing, bound proteins were eluted with either 250 mM imidazole or 50 mM biotin, respectively. Eluted fractions were analyzed by SDS-PAGE and Western blotting/in-gel fluorescence.

Cryo-EM grid vitrification and data collection

For GluK3 D759GUBP310+spermine, prior to vitrification, 0.5 mM UBP310 and 1 mM spermine were added to the purified protein and incubated for 20 min on ice to trap the GluK3 (D759G) receptor in a closed state. Quantifoil 0.6 Au 300 mesh grids were glow-discharged for 90 s at 25 mA. Three microliters of protein at 2.5 mg/ml was applied to the grid once, followed by blotting for 3.5 s at 100% humidity and 4 °C, and then vitrified by plunging into ethane cooled by liquid nitrogen using a Vitrobot. The grids were sent to the European Synchrotron Radiation Facility for data collection at CM01.

The grids were loaded onto Titan Krios, accelerated at 300 kV, and a total of 8969 multi-frame movies (40 frames/ movie) were collected using a K2 detector with a physical pixel size of 1.052 μm in counting mode, with a total dose of 43.72 e/Å2. Movies were motion-corrected using UCSF Motioncorr247. The contrast transfer function of the aligned micrographs was estimated using CTFFIND4, and particles were manually picked from a few micrographs in cryoSPARCv248,49. The picked particles were extracted and subjected to 2D classification. Class averages with desirable features were selected and used to train Topaz50 for automated particle picking in cryoSPARC v3.1. 151525 particles were picked initially; however, after several 2D and 3D classification rounds, 56089 particles were used for final 3D reconstruction. The parameters used for data collection, processing, refinement, and validation are tabulated in Table 3.

Table 3 CryoEM data collection, refinement, and validation statistics for GluK3 D759GUBP310+spermine
Full size table

Model building

The structure of GluK3 complexed with UBP310 (PDB 6jfz) was used as the initial model to be fitted into the 3D EM density map in UCSF Chimera51. Real-space refinement was performed using Phenix, and the Ramachandran outliers were fixed in Coot52,53. To minimize the clash score further, individual chains of the refined model were used to split the EM map in UCSF Chimera, and the models of individual chains/subunits were subjected to flexible fitting and refinement using an online tool called Namdinator and Phenix (Namdinator- Automatic Molecular Dynamics Flexible fitting of structural models into cryoEM and crystallography experimental maps). The individual models for each subunit were combined, and the S2-M4 linkers were manually built and fitted into the density map using Coot52. After combining the chains, rigid body refinement in Phenix was performed to yield the final model for the GluK3D759G receptor in the presence of UBP310 and spermine.

Statistics and reproducibility

Comparisons between conditions were made by paired and unpaired t-test (two-tailed) wherever applicable. Statistical analysis was carried out in Prism, version 8.01 (GraphPad software). P values < 0.05 were considered statistically significant and are reported (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). All experiments were repeated independently with consistent results, including those for which statistical analysis was not applicable.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.