Mather et al. propose that arousal modulates attention through a norepinephrine–glutamate feedback loop in local circuits. Here, I suggest a specific circuit where this mechanism may be in operation: the granule cell–mossy cell loop in the hippocampus.

It is commonly proposed that the CA3 region of hippocampus forms an auto-associative memory store for short- and medium-term memories (Gardner-Medwin Reference Gardner-Medwin1976; Hopfield Reference Hopfield1982; Levy & Steward Reference Levy and Steward1979; Marr Reference Marr1971; McNaughton & Morris Reference McNaughton and Morris1987; Rolls Reference Rolls, Durbin, Miall and Mitchison1989; Treves & Rolls Reference Treves and Rolls1992). Here, pictures, memories, in the form of patterns of activity in the entorhinal cortex, feed forward along the perforant pathway to CA3, activating a sparse subset of the CA3 pyramidal cells. Plasticity in the synapses of the recurrent network in CA3 and in the perforant pathway synapses onto CA3 neurons fixes the memory so that it can be recalled: If a part of the same pattern of activity occurs in entorhinal cortex, the corresponding part pattern is activated in CA3 and it is then completed by auto-associative dynamics.

Pattern collision, where two similar memories are confused during pattern completion, is a problem in auto-associative networks, particularly if they are required to rapidly store memories with only a small number of presentations. It is likely that the hippocampus has a mechanism to avoid or reduce pattern collision: the hippocampus stores rapidly acquired memories, and it is important that similar but distinct memories can be distinguished during recall.

It has been proposed that the role of the dentate gyrus is to separate patterns and thereby reduce collisions (Gilbert et al. Reference Gilbert, Kesner and Lee2001; Leutgeb et al. Reference Leutgeb, Leutgeb, Moser and Moser2007; McHugh et al. Reference McHugh, Jones, Quinn, Balthasar, Coppari, Elmquist, Lowell, Fanselow, Wilson and Tonegawa2007; O'Reilly & McClelland Reference O'Reilly and McClelland1994; Treves & Rolls Reference Treves and Rolls1992). In addition to CA3 neurons, the perforant pathway connects to the granule layer in dentate gyrus. The granule layer of dentate gyrus is, in turn, connected to CA3 along the mossy fibers. This means that the entorhinal cortex is connected to CA3 directly, along the perforant pathway, and indirectly, via dentate gyrus. In the specific version of dentate gyrus pattern separation proposed by O'Reilly and McClelland (Reference O'Reilly and McClelland1994), there is local k-winner-take all dynamics between cells in dentate gyrus, and the consequence of this is that only a random subset of the cells receiving input from entorhinal cortex become active. This activity is fed forward along the mossy fibers to CA3 and, in turn, excites a random subset of those cells in CA3 that receive input from entorhinal cortex. This randomization separates the patterns that are then learned in the CA3 auto-associative network.

There is experimental evidence (McHugh et al. Reference McHugh, Jones, Quinn, Balthasar, Coppari, Elmquist, Lowell, Fanselow, Wilson and Tonegawa2007) that the dentate gyrus is important for pattern separation and that the adult neurogenesis of dentate gyrus granule cells, which may support the randomization, is linked to pattern separation (Altman Reference Altman1963; Bayer et al. Reference Bayer, Yackel and Puri1982; Clelland et al. Reference Clelland, Choi, Romberg, Clemenson, Fragniere, Tyers, Jessberger, Saksida, Barker, Gage and Bussey2009; Sahay et al. Reference Sahay, Scobie, Hill, O'Carroll, Kheirbek, Burghardt, Fenton, Dranovsky and Hen2011). However, it seems unlikely that pattern separation is the only role of the dentate gyrus; for a start, pattern separation on its own seems a modest role for such a substantial brain region. Beyond this, pattern separation does not explain either the hilar region or the role of norepinephrine in the dentate gyrus.

The hilar region lies between dentate gyrus and CA3. As the mossy fibers run through the hilar region they form en passant connections with the mossy cells (Amaral Reference Amaral1978; Scharfman & Myers 2013). These are large excitatory cells whose proximal dendrites are covered in mossy-looking spines. The mossy cells, in turn, have a substantial backprojection that extends along the longitudinal axis of the dentate gyrus (Amaral & Witter Reference Amaral and Witter1989; Amaral et al. Reference Amaral, Scharfman and Lavenex2007) and connects to both granule cells and inhibitory interneurons (Scharfman Reference Scharfman1994; Reference Scharfman1995).

This two-layer structure seems more elaborate than a simple randomizing k-winner-takes-all network would require; random subselection from a pattern could be achieved by local excitatory–inhibitory dynamics within the dentate gyrus itself. However, the two-layer structure would make sense if the role of the dentate gyrus encompassed memory selection as well as pattern separation. As pointed out by Koch et al. (Koch & Ullman Reference Koch and Ullman1984; Reference Koch, Ullman and Vaina1987; Olshausen et al. Reference Olshausen, Anderson and Van Essen1993), a single layer winner-takes-all network in which competition occurs across the whole network requires considerable interneuronal connectivity. This issue is resolved by having more than one layer; in the first layer, competition is restricted to subregions, and a champion emerges from each subregion to compete in the next layer where the competition between subregions occurs. In short, I suggest here that, in addition to separating patterns, the winner-take-all dynamics in the dentate gyrus also compares the salience of different aspects of its input and that this selection gates and refines the storage of memories in CA3. The role of the hilar region is to facilitate this comparison.

The locus coeruleus projects to the dentate gyrus, which contains β-adrenergic receptors (Berridge & Waterhouse Reference Berridge and Waterhouse2003; Harley Reference Harley2007). Norepinephrine release in response to novelty during exploration enhances excitability in the dentate gyrus (Kitchigina et al. Reference Kitchigina, Vankov, Harley and Sara1997); in fact, the activity of both interneurons (Nitz & McNaughton Reference Nitz and McNaughton2004) and excitatory neurons (Dahl & Winson Reference Dahl and Winson1985; Neuman & Harley Reference Neuman and Harley1983) in dentate gyrus show norepinephrine-promoted increase in response to novelty. Furthermore, it has been reported that in hippocampus, glutamate causes enhanced norepinephrine release (Pittaluga & Raiteri Reference Pittaluga and Raiteri1990; Raiteri et al. Reference Raiteri, Garrone and Pittaluga1992), an effect that is most marked in the dentate gyrus (Andrés et al. Reference Andrés, Bustos and Gysling1993). Conversely, norepinephrine in dentate gyrus, but not in other hippocampal regions, potentiates the release of glutamate (Lynch & Bliss Reference Lynch and Bliss1986). The role of norepinephrine in dentate gyrus seems somewhat mysterious if the role of the dentate gyrus is restricted to pattern separation. However, if, as proposed here, the dentate gyrus also performs memory selection, then the norepinephrine–glutamate mechanism for modulating memory selectivity described by Mather et al. becomes the missing clue that could explain the role of norepinephrine in dentate gyrus.