(A) In response to an odor, a sparse ensemble of Kenyon cells provides excitatory synaptic input to MBONs (black arrows; Hige et al., unpublished) (Cassenaer and Laurent, 2007). Glutamatergic (Glu; green), GABAergic (GABA; blue) and cholinergic (Ach; orange) MBONs all receive KC input; the names of MBONs are based on the lobe compartment where their dendrites arborize. CsChrimson activation of some glutamatergic MBONs can be repulsive, whereas activation of GABA or cholinergic MBONs can be attractive (color coded as indicated by the scale at bottom right) (Figure 2C). We often only observed significant behavioral effect with combinations of cell types (indicated by dashed lines grouping multiple cell types). MBON-γ1pedc>α/β, and glutamatergic MBONs, MBON-β1>α and MBON-γ4>γ1γ2, have synaptic terminals inside the compartments of MB lobes (Aso et al., 2014). While the microcircuits within each MB compartment remain to be elucidated, our light level anatomical studies have enumerated the cell types present in each compartment. The α2 and α3 compartments contain the dendrites of cholinergic MBONs and are targeted by both the GABAergic MBON-γ1pedc>α/β and the glutamatergic MBON-β1>α. These MBONs cover only a small fraction of volume in their target compartments (see Figure 6—figure supplement 1); since they could contact only a fraction of Kenyon cells, we propose that they target MBONs directly. Here we hypothesize that glutamate is inhibitory to cholinergic and GABAergic MBONs, and GABA is inhibitory to glutamatergic MBONs (inhibitory connections are indicated by circles). The thickness of lines and size of their endings are meant to indicate activity levels. (B) PPL1- γ1pedc DANs play a major role to mediate punishment signals to the MB for formation of aversive memory together with minor contribution from other DANs including PPL1-γ2α′1 (Aso et al., 2012). If an aversive memory is formed by the simultaneous presentation of an odor and punishment and results in a synaptic depression of KC terminals by dopamine (represented by red dashed circles), the response of the GABAergic MBONs to the CS+ would be depressed (see text). Reduced GABAergic inhibitory input would then increase the CS+ response of glutamatergic MBONs, whereas the CS+ response of cholinergic MBONs would be reduced because of dis-inhibition of the inhibitory glutamatergic MBONs. The end result is enhanced activity of aversion-mediating glutamatergic MBONs together with the reduced activity of attraction-mediating GABAergic and cholinergic MBONs in response to CS+. The CS- is represented by a different set of KCs whose synaptic connections to the MBONs would not be expected to be modified by training and so the responses of the ensemble of MBONs to the CS- would remain balanced. The change in response of the MBONs to the CS+, relative to their unchanged response to the CS-, biases choice toward the CS+ (see diagram in panel D). This model is consistent with the essential role of MBON-γ1pedc>α/β in aversive memory. Also, cholinergic MBONs in V2 cluster have been shown to reduce their response to an odor after olfactory conditioning with electric shock (Séjourné et al., Nat. Neurosci., 2011), although we detected their requirement only for long-term memory, but not for 2 hr memory. This model predicts a role for glutamatergic MBONs, but we did not observe significant effect by blocking subsets of glutamtergic MBONs (Figures 6 and 8). Thus, to test this model, it will be necessary to block broader sets of glutamatergic MBON cell types by using combinations of split-GAL4 drivers. (C) In contrast to aversive memory in which one type of DAN (PPL1-γ1pedc) plays a major role in memory formation, reward signals are mediated by a distributed set of PAM cluster DANs that innervate the compartments of glutamatergic MBONs (Yamagata et al., in press) (Burke et al., 2012; Liu et al., 2012; Perisse et al., 2013). If an appetitive memory is formed by synaptic depression of KC terminals in response to dopamine release, the response of the glutamatergic MBONs to the CS + would be depressed. The resultant reduced glutamatergic inhibitory input to the GABAergic and cholinergic MBONs would increase their response to the CS+. In turn, increased GABAergic input to glutamatergic MBONs may further amplify and stabilize the initial effects of plasticity. The end result would be reduced activity of aversion-mediating glutamatergic MBONs together with the increased activity of attraction-mediating GABAergic and cholinergic MBONs in response to CS+. This model is consistent with requirement of glutamatergic MBONs: blocking the M4/M6 cluster MBONs (MBON-γ5β′2a, MBON-β′2mp and MBON-β′2mp_bilateral; Table 1) by MB011B resulted in memory impairment for all three appetitive memory assays (Figures 7, 9 and 11). Blocking the MBON-γ4>γ1γ2 and MBON-β1>α by MB434B resulted in memory impairment in two of three assays (Figures 9 and 11). While we did not detect a requirement for the GABAergic MBON-γ1ped>α/β in appetitive memory, previous study have shown that dopamine input to the γ1 and pedc suppresses expression of appetitive memory in fed flies (Krashes et al., 2009), indicating some role of MBON-γ1ped>α/β in appetitive memory. Blocking cholinergic MBONs in the V3/V4 cluster (MBON-γ2α′1, MBON-α′2 and MBON-α3) resulted in memory impairment in appetitive odor memories (Figures 7 and 11) (Placais et al., 2013) but not in appetitive visual memory (Figure 9). Some of driver lines for the cholinergic MBONs in the V2 cluster showed impairment of appetitive memory in all three assays (Figures 7, 9 and 11), although our data did not allow mapping to the resolution of specific cell types due to inconsistent results obtained using other lines. (D) In the matrix shown, the number of circles represents the activity levels of MBONs in response to the CS+ odor (the odor that is paired with the unconditioned stimulus during conditioning) and the CS− odor (a control odor). In untrained flies, the activities of MBONs for opposing effects are balanced. Dopamine modulation breaks this balance to bias the choice between CS+ and CS−.