An appealing theoretical aspect of global multiplicative synaptic scaling is the preservation of the pattern of relative differences in synaptic weights established by Hebbian forms of synaptic plasticity that is postulated to encode information [9
]. However, while uniform synaptic scaling has been reproducibly observed in young neurons under appropriate conditions, older neurons (here defined as those beyond the period of bulk synaptogenesis, for example, >DIV21 or in the adult animal) from a variety of preparations do not show scaling, even with global activity manipulations [27
]. The occurrence of multiplicative scaling only during the period of peak synaptogenesis (and not in older neurons) suggests that this mechanism may actually be more relevant to synapse formation rather than information processing per se.
Instead, TTX applied to older neurons elicits nonmultiplicative increases in mEPSC amplitudes [56
], as well as elevated mEPSC frequency (e.g., [41
]; see for others). A perplexing question that then arises is that if synapse strength is affected in a nonuniform way, how can homeostatic adjustments coexist with Hebbian information encoding? One proposal for allowing the coexistence of Hebbian and homeostatic mechanisms is if the former is implemented by dynamically moving the set point of the latter [8
], in much the same way that a thermostat can be turned up or down, but still remains under feedback control. However, this mechanism does not explain the nonmultiplicative HSP in older neurons. The basis of this HSP in mature neurons remains unknown, but by definition a nonmultiplicative process implies that certain synapses are affected differentially, and in mature neurons HSP has indeed been shown to influence larger synapses disproportionately [24
]. The implication of these results is that, in older neurons, some synapses retain higher capacity to generate strong homeostatic responses, while others may become relatively insensitive to chronic changes in activity. We note that the latter population would be ideally suited to durable and persistent information encoding. We speculate that this hypothetical division of plasticity labor would nicely allow homeostatic adjustment without interference with Hebbian plasticity, but such a mechanism remains to be identified and described.
Consistent with the notion that older neurons have populations of synapses that may be resistant to homeostatic adjustment, blocking presynaptic neurotransmitter release at single synapses with tetanus toxin transfection in mature hippocampal neurons did not cause changes in AMPAR-mediated currents at contacting postsynaptic sites but did cause changes in NMDAR subunit composition in an interesting form of metaplasticity or the “plasticity of plasticity” [79
]. In older neurons, metaplasticity may provide an attractive alternative (or additional) strategy for restraining the capacity of Hebbian plasticity without interfering with synaptic weighting [7
]. Alternatively, changes in presynaptic release probability may allow for homeostatic adjustments without altering postsynaptic information encoding. Indeed, in the intact adult hippocampus, CA1 synapses do not show mEPSC amplitude changes in response to TTX but only increased frequency [56
In vivo, network stability may also arise as a consequence of the specific arrangement of connectivity and not merely the individual synaptic strengths. For instance, chronic inactivity in mature organotypic hippocampal slices induced upregulation of synaptic efficacy in a manner which reflected the underlying computations of the network. Within the hippocampal trisynaptic circuit, CA3 “throughput” synapses were upregulated in response to inactivity, while recurrent synapses were downregulated [54
]. It is therefore possible that, in functional circuits, certain synaptic interfaces are a designated homeostatic locus. Similar synapse-specific adaptations have been detected in the visual system, and interestingly the locus of the homeostatic adaptation appeared to change with development. Visual deprivation induced selective homeostatic adaptation in layer II/III neurons in adult visual cortex, while inducing selective layer IV adaptation in developing neurons [57
]. These results suggest not only that multiple HSP mechanisms exist in vivo [27
] but also that specific cell types may differentially mediate HSP and that the computations of the network at different developmental time points can alter the locus of homeostatic adaptation.