Integration of Parallel Opposing Memories Underlies Memory Extinction

Abstract

Accurately predicting an outcome requires that animals learn supporting and conflicting evidence from sequential experience. In mammals and invertebrates, learned fear responses can be suppressed by experiencing predictive cues without punishment, a process called memory extinction. Here, we show that extinction of aversive memories in Drosophila requires specific dopaminergic neurons, which indicate that omission of punishment is remembered as a positive experience. Functional imaging revealed co-existence of intracellular calcium traces in different places in the mushroom body output neuron network for both the original aversive memory and a new appetitive extinction memory. Light and ultrastructural anatomy are consistent with parallel competing memories being combined within mushroom body output neurons that direct avoidance. Indeed, extinction-evoked plasticity in a pair of these neurons neutralizes the potentiated odor response imposed in the network by aversive learning. Therefore, flies track the accuracy of learned expectations by accumulating and integrating memories of conflicting events.

Keywords: Drosophila; competition; connectomics; dopamine; extinction; memory; neural circuit; neural plasticity; parallel memory.

Graphical abstract

Figure S1

Aversive Memory Extinction Requires PAM Cluster Dopamine Neurons, Related to Figure 1 (A) Pre-exposing naive flies to high (10⁻³), but not low (10⁻⁴-10⁻⁶) concentration of odor, biases subsequent choice behavior toward avoidance of that odor in an ITI dependent manner. Low odor concentrations were therefore used in all behavioral experiments in this study. (B) Learning performance is similar with low (10⁻⁶) and high (10⁻³) odor concentrations. (C) Retraining after extinction reverses the reduction in learned avoidance behavior (left) and leads to more robust 90 min aversive memory (right). (D) Permissive temperature control experiment for Figure 1D. All the relevant groups show normal extinction when performance is measured 60 min after training. (E) Blocking R58E02-GAL4 dopamine neurons during the retrieval of an extinguished memory does not impair test performance. 60 min performance of R58E02-GAL4; UAS-Shi^ts1 flies not statistically different from controls. Asterisks, significant difference between groups of same genotype. Data, mean ± SEM. All individual data points displayed as dots.

Figure 1

Extinction of Aversive Memory Requires PAM Dopamine Neurons (A) Top: protocol. Bottom: two (15 min ITI) or five (1 min ITI) CS+ re-exposures induces extinction. (B) Top: protocol. Bottom: CS+, but not CS−, re-exposure induces extinction. (C) Left: protocol with temperature shifting (dashed line) and R58E02-GAL4 DANs schematic. Right: blocking R58E02 neurons with UAS-Shi^ts1 during CS+ re-exposure abolishes aversive memory extinction. (D) Left: protocol and MB504B-GAL4 DANs schematic. Right: blocking MB504B neurons during CS+ or CS− re-exposure does not alter extinction. Asterisks denote significant differences. Data are represented as the mean ± SEM; individual data points are displayed as dots. See also Figure S1 and Table S1.

Figure 2

M4β′ and M6 MBONs Drive γ5 Dopamine Neurons and Memory Extinction (A) M4β′ and M6 neurons were activated with CsChrimson and GCaMP6f Ca²⁺ responses measured in presynapses of specific PAM DANs, identified by compartment innervation in the MB lobes. Insets: maximum projections of imaging planes with pseudo-colored GCaMP signals and γ4, γ5, β′2p, and β′2 m regions of interest (ROIs). Controls express GCaMP6f, but not CsChrimson. (B) M4β′/M6 activation evoked significant signals in γ5 PAM DANs. (C–E) M4β′/M6 activation did not evoke significant signals in γ4 PAM DANs (C), whereas it produced subtle inhibition and rebound in PAM β′2p (D) and PAM β′2 m (E) DANs. Arrows, time points, before and after light stimulation (red box), used for quantification. Paired measurements connected (green lines) and mean response (gray bar). (F) Top: protocol with temperature shifting. Bottom: blocking VT1211-GAL4 M4β′/M6 with UAS-Shi^ts1 during CS+ re-exposure (left) reduces extinction. Blocking M4β′/M6 during CS− re-exposure abolishes odor avoidance behavior (right). Asterisk denotes significant difference. ns, no significant difference. Asterisks in (A)–(E) indicate a significant difference between pre- and post-activation responses. Data in (F) mean ± SEM. Dots represent individual data points. See also Figure S2 and Table S1.

Figure S2

Mushroom Body Output Neurons Drive γ5 Dopamine Neurons and Memory Extinction, Related to Figure 2 (A) Control functional connectivity experiment using flies which express GCaMP6f in PAM DANs but not CsChrimson in M4β′/M6 neurons, light alone does not evoke a Ca²⁺ response in PAM DANs innervating γ5. ns, no significant difference. (B) Blocking VT1211-GAL4 labeled M4β′/M6 neurons with UAS-Shi^ts1 for 45 min after aversive conditioning does not alter 60 min memory performance. (C) Permissive temperature control experiment for Figure 2F. All groups show comparable avoidance behavior after extinction (left) or after CS- re-exposure (right). Data are mean ± SEM and all individual data points are displayed (dots).

Figure 3

Parallel Memory Traces Form When Aversive Memory Is Extinguished (A) Imaging plane in MVP2 dendritic field, training and imaging protocol under the microscope. Aversive conditioning significantly reduces CS+ responses in MVP2. (B and C) No differences evident in M4β′ (B) or M6 (C) dendrites after aversive conditioning. (D) Extinction protocol. Training induced reduction in CS+ response in MVP2 remains after extinction. (E) Odor responses in M4β′ dendrites unchanged following extinction. (F) Extinction induces relative decrease in the CS+ response in M6 dendrites. Odor-evoked activity traces, mean (solid line) and SEM. (shadow). Black line, 5 s odor presentation. Paired measurements from individual flies shown as black (CS+ response < CS− response) or white (CS+ response > CS− response) dots. Asterisks, significant difference between averaged CS+ and the CS− responses. ns, no significant difference. See also Figures S3 and S6 and Table S1.

Figure S3

Parallel Memory Traces Form When Aversive Memory Is Extinguished, Related to Figure 3 (A and B) Complementary experiment for Figures 3C and 3F. (A) Training protocol under the microscope using 3-octanol (OCT) as CS+. Aversive conditioning does not significantly change the CS+ odor response in M6 dendrites. (B) Extinction protocol under the microscope. CS+ odor responses in M6 dendrites are significantly reduced after extinction. (C and D) Complementary experiment for Figures 3A and 3B. (C) Aversive conditioning significantly reduces CS+ odor responses in MVP2 dendrites. (D) The training induced reduction in CS+ odor (OCT) response in the MVP2 dendrites remains after extinction. (E–H) Repeating experiments in Figures 3C, 3F, S3A, and S3B with the VT1211-GAL4 driver confirms the findings: (E and G) aversive conditioning does not change odor responses in M6 dendrites. (F and H) However, aversive memory extinction leads to reduced CS+ response in M6 dendrites. (I and J) Complementary experiment for Figures 3B and 3E. (I) No differences evident in odor-evoked responses in M4β′ dendrites after aversive conditioning. (J) Odor responses measured in M4β′ dendrites are unchanged following extinction of aversive memory. (K–R) Mock conditioning, exposing flies to the same odor training regime without electric shock, does not change odor responses measured in (K and N) MVP2 or (O-R) M6 dendrites. Odor-evoked activity traces show mean (solid line) with SEM. (shadow). Black line represents 5 s odor presentation. Paired measurements from individual flies displayed as black (CS+ response < CS- response) or white (CS+ response > CS- response) dots. Asterisks, significant difference between averaged CS- and the CS+ responses.

Figure 4

MVP2 Neurons Connect in Different Ways to M4β′ and M6 Neurons (A) 3D view of right-brain hemisphere MVP2, M4β′, and M6 neurons from EM tracing. MVP2, M4β′, and M6 neurons have ipsi- and contralateral processes in MB lobes (black outline). Scale bar, ∼20 μm. Dorsal, ventral, medial, and lateral directions indicated. See also Video S1. (B) Confocal projections of MVP2 (cyan), M4β′ (orange), and M6 (magenta), where they intersect in MB lobes. Top row: MVP2 processes intermingle with M4β′ peripheral dendrites in β′2 (white arrows). Bottom row: large diameter MVP2 axon branches (cyan) overlay M6 dendrites in γ5 (white arrows). Scale bar, 10 μm. See also Videos S2 and S3. (C) EM tracing of MVP2 inputs to M4β′ dendrites. MVP2 presynapses (red dots) opposing M4β′ postsynapses. M4β′ axonlets, presynaptic processes extending from dendritic fields are marked (green shades). Two dendritic branches project into crepine (arrows). Scale bar, ∼5 μm. (D) Dendrogram of M4β′ neuron showing postsynapses opposing MVP2 inputs (red dots), KC inputs (blue), presynaptic output (green). Neurite length not preserved. Main axon branches, arrows. Asterisks, axonlets. (E) EM tracing of MVP2 input to M6 dendrites. Annotated as in C. MVP2 inputs (red) often cluster along M6 major neurites. Scale bar, ∼2.5 μm. (F) Dendrogram of M6. MVP2 input (red), KC input (blue), presynaptic output (green). Two primary axon branches (black arrows). Contralateral axon (gray arrow). See also Figure S4.

Figure S4

MVP2 Neurons Connect in Different Ways to M4β′ and M6 Neurons, Related to Figure 4 (A) Alternative 3D view of the projections of the fly’s right brain hemisphere MVP2, M4β′ and M6 neurons from EM tracing. MVP2 processes in the horizontal MB lobe, innervate β′2 and γ5 compartments, occupied by dendrites of M4β′ and M6 neurons, respectively. Scale ∼20μm. Dorsal, ventral, medial and lateral directions indicated. (B) Analysis of ultrastructural connectivity between MVP2 neurons from right and left hemispheres with the right hand M4β′ and right and left M6 neurons. Numbers of synapses between respective neurons indicated on connections between boxes. (C) Analysis of dendritic field of left M6(L) neuron confirms that MVP2 inputs (red dots) localize near the root of dendrites. Scale ∼2.5 μm. (D) Dendrogram of placement of MVP2 inputs to M6(L). MVP2 input (red), likely-KC input (blue), presynaptic output (green). Two primary axon branches indicated (black arrows). (E) Quantification of localization of MVP2 input to M4β′ and M6(R) and M6(L) neurons with respect to their distance to root of the dendritic field. Non-MVP2 inputs, assumed to be mostly KCs, onto M4β′, M6(R) and M6(L) neurons have a Gaussian or bimodal Gaussian distribution, spread over the dendritic field. MVP2 inputs to M4β′ are more distally localized than MVP2 inputs to M6(R) and M6(L) neurons.

Figure 5

Ultrastructure of MVP2 Synapses onto M4β′ and M6 Neurons (A) 3D view of EM tracing of MVP2 presynapses within M4β′ dendritic field (same as Figure 4C). Most MVP2 presynapses are on bouton-like structures. (B) Inset, area modeled in 3D reconstruction of MVP2 bouton (blue) and corresponding presynapses (red) making inputs to different M4β′ dendrites (orange). Scale bar, 1 μm. Gray lines (1–4), locations of EM data shown in (C). (C) 1, EM image of MVP2 synapse onto unrelated neuron (white arrow). 2–4, MVP2 to M4β′ synapses. This MVP2 bouton also inputs to other neurons. Synaptic cleft (red). Scale bar, 1 μm. (D) 3D view of MVP2 presynapses within M6 dendrite (same as Figure 4E). MVP2 makes en passant synapses onto spines in M6 dendritic field. Scale bar, 1 μm. (E) Inset, area modeled in 3D reconstruction of en passant MVP2 (blue) presynapses (red) onto spine-like twigs (asterisk) of M6 (magenta). Gray lines (1–4), locations of EM sections in (F). (F) 1, presynapse (red) from MVP2 (R) (blue) onto postsynapse of M6 spine-like twig (magenta). 2, section between synapses. 3, white arrow MVP2(L) presynaptic site (red). 4, synaptic cleft (red) of same presynapse extending into two postsynaptic sites (inputs) onto M6 spine-like twig (magenta). Scale bar, 500 nm. See also Videos S4 (MVP2-M4β′) and S5 (MVP2-M6).

Figure 6

Aversive and Extinction Memories Are Integrated in M6 Neurons (A) Imaging plane and protocol. Aversive conditioning increases CS+ response in the axon of M4β′/M6 MBONs. (B and C) Potentiated response to CS+ evident in M4β′ (B) and M6 (C). (D) Extinction protocol. Extinction nullifies training-induced increase in CS+ response in M4β′/M6 axons. (E) Training-induced potentiation of CS+ response in M4β′ axon survives extinction. (F) Extinction nullifies training-induced increased CS+ response in M6 axon. Odor-evoked activity traces show mean (line) with SEM. (shadow). Black line, 5-s odor during imaging phase of the experiment. Paired measurements are the same as those used in Figure 3. Asterisks, significant difference between CS− and CS+ responses. See also Figures S5 and S7 and Table S1.

Figure S5

Aversive and Extinction Memories Are Integrated within M6 Neurons, Related to Figure 6 (A and B) Complementary experiment to Figures 6A and 6D. (A) Imaging plane and training protocol under the microscope. CS+ is OCT. Aversive conditioning increases CS+ odor response of axonal segment of M4β′ and M6 neurons. (B) Extinction protocol under the microscope. Aversive memory extinction nullifies the training-induced increase in CS+ odor response measured in axonal segment of M4β′/M6 neurons. (C and D) Complementary experiment to Figures 6C and 6F. (C) The potentiated response to the CS+ after aversive conditioning is evident in the axonal segment of M6 neuron. (D) Consistent with Figure 6F, extinction of aversive memory for OCT nullifies the training-induced increased CS+ odor response in the M6 axon. (E–H) Mock conditioning with OCT does not change odor responses measured in axonal segments of M4β′ and M6 neurons. Odor-evoked activity traces show mean (solid line) with SEM. (shadow). Black line, 5 s odor presentation. Paired measurements from individual flies displayed either as black (CS+ response < CS- response) or white (CS+ response > CS- response) dots. Asterisks, significant difference between averaged CS- and the CS+ response.

Figure S6

Parallel Memory Traces Form When Aversive Memory Is Extinguished, Related to Figures 3 and S3 (A–X) All imaging traces for odor responses to the CS+, the CS- (either OCT, blue, or MCH, red) or the novel odor (IAA, orange) for the experiments in the order as they are depicted in Figures 3 and S3. Individual traces (gray), the mean (colored solid line) and the SEM (shadow) are displayed. Black line represents 5 s odor presentation during the imaging phase of the experiment.

Figure S7

Aversive and Extinction Memories Are Integrated within the M6 Neurons, Related to Figures 6 and S6 (A–N) All imaging traces for odor responses measured in M6 axons to the CS+, the CS- (either OCT or MCH) or the novel odor (IAA) for the experiments in the order as they are depicted in Figures 6 and S6. Individual traces (gray), the mean (colored solid line) and the SEM (shadow) are displayed. Black line represents 5 s odor presentation during the imaging phase of the experiment.

Figure 7

Model of Extinction: Aversive Memory Expression Is Limited by Competition with a Parallel Extinction Memory of Opposite Valence (A) Individual DANs from the PPL1 and PAM clusters innervate distinct mushroom body lobe compartments. PPL1-DANs (red) provide teaching signals during aversive conditioning and PAM-DANs (green) for appetitive conditioning. Each compartment, innervated by a particular DAN also houses dendrites of a corresponding MBON, which are GABAergic (blue), glutamatergic (magenta), or cholinergic (not shown). MBONs receive excitatory acetylcholine from odor coding KCs (black). Terminals of PPL1 DANs overlap dendrites of MBONs promoting approach behavior (blue), whereas PAM DANs overlap MBONs directing avoidance (magenta). MBONs drawn are valence-coding MBONs described to harbor traces of aversive or appetitive memory (Owald et al., 2015, Perisse et al., 2016). (B) When a naive fly detects neutral odor, odor-specific KCs (black) drive an equally weighted network of approach and avoidance promoting MBONs. This balanced network configuration does not promote directed behavior. (C) During aversive conditioning, CS+ induced activity in KCs and downstream MBONs coincides with activity of PPL1 DANs, leading to compartment restricted synaptic depression between odor-activated KCs and respective MBON. (D) Following aversive conditioning, CS+ drive to approach MBONs is reduced (smaller triangle) and as result of reduced odor-specific MVP2-mediated feedforward inhibition, CS+ drive to avoidance promoting M4β′/M6 is also potentiated. (E) During extinction, learned configuration of the MBON network favors CS+ activation of avoidance promoting MBONs, which, in turn, drives appetitively reinforcing γ5 PAM DANs. Coincidence of CS+ during extinction and γ5 DAN activity depresses odor-activated KC synapses onto M6 MBONs. (F) After extinction, reduced CS+ drive to avoidance coding M6 (smaller triangle) partially compensates for the network potentiation of M6 neuron response induced during initial aversive training.