Since the discovery of TARPs a decade ago, numerous studies have established that these auxiliary subunits control the trafficking and gating of AMPARs, and that the magnitude of this effect depends on the specific TARP subtype (Cho et al., 2007
; Kott et al., 2007
; Milstein et al., 2007
). Might the number of TARP molecules that associate with individual AMPAR channels also regulate gating? Although previous data raise this possibility (Milstein et al., 2007
), resolving TARP/AMPAR stoichiometry has been surprisingly difficult. We have genetically linked TARPs to AMPARs and have thus been able to identify the number of TARPs associated with native AMPARs.
By expressing AMPAR subunits covalently linked to TARPs in HEK cells we have compared the properties of AMPARs containing zero or four TARPs and receptors with an intermediate stoichiometry. We show that kainate efficacy, previously shown to increase with TARP association (Tomita et al., 2005
; Turetsky et al., 2005
), varies with TARP/AMPAR stoichiometry and can be used to distinguish between AMPARs that are saturated or unsaturated by TARP expression. The functional effects of TARPs γ-2 and γ-8 on fused GluA1 and GluA2 were similar, and both varied with TARP/AMPAR stoichiometry. These results not only establish that AMPARs can assemble with different numbers of TARPs, but also provide valuable quantitative data that can be compared to native AMPARs in neurons. Although not addressed in this study, it would not be surprising if the magnitude of other modulatory effects of TARPs, such as the sensitivity of Ca2+
-permeable AMPARs to block by intracellular polyamines (Soto et al., 2007
), and the trafficking of AMPARs to synapses, might be similarly regulated by stoichiometry.
When we expressed heteromeric receptors in which only one of the subunits was tethered to a TARP we consistently obtained receptors with a kainate efficacy that was roughly intermediate between that observed with zero-TARP and four TARP receptors. This finding is consistent with the model in which AMPARs are assembled with a stoichiometry of two GluA1 and two GluA2 subunits (Mansour et al., 2001
). If, however, subunits can randomly assemble with a variable subunit stoichiometry (Washburn et al., 1997
), then the intermediate kainate efficacy could represent a mixed population of heteromers with receptors containing two TARPs predominating. Another possible interpretation of our results is that the kainate efficacy actually saturates when less than four TARPs are bound to the receptor. Although this would not alter the conclusion that stoichiometry can vary, it would question the quantitative conclusions. However, a simple set of calculations that predicts the kainate efficacies of AMPARs that are produced by varying models of channel assembly (Figure S3B
) supports the scenario in which heterodimer formation is preferred and the ratio saturates at four TARPs.
Remarkably, we find that AMPARs in CA1 pyramidal neurons are normally associated with four molecules of TARP γ-8. When the expression level of γ-8 is reduced, the number of TARPs associated with each AMPAR is reduced. At this point we can only speculate as to what the precise TARP/AMPAR stoichiometry is in γ-8 (−/−) and γ-8 (+/−) mice. If we assume that at least one TARP must be associated with AMPARs in order for them to be efficiently trafficked to the cell surface, then we would expect all surface receptors to contain at least one TARP; indeed, the observed kainate efficacy of γ-8 (−/−) pyramidal neurons was between that expected for zero and two TARPs, indicating that perhaps it is just one TARP. In this scenario, TARPs γ-2 and γ-3, which are also expressed in pyramidal neurons (Lein et al., 2007
; Tomita et al., 2003
), are associated with the remaining AMPARs in γ-8 (−/−) neurons. In addition, we demonstrate in CA1 neurons the time course of AMPAR deactivation and desensitization varies with γ-8 expression level, consistent with the change of TARP/AMPAR stoichiometry.
Furthermore, we find that TARP/AMPAR stoichiometry differs between distinct neuronal cell types. Unlike CA1 pyramidal neurons, hippocampal dentate gyrus granule cells are not saturated with four TARPs, but instead appear to express a mixed population of AMPARs containing fewer than four TARPs. This is also consistent with the data we previously acquired from cerebellar granule neurons (Milstein et al., 2007
), and reinforces the notion that differences in the expression level of different TARP subtypes diversifies the functional properties of neuronal AMPARs, not only through subtype-specific regulation, but also through variable TARP/AMPAR stoichiometries.