Biochemical evidence accumulates across neurons to drive a network-level eruption

Abstract

Neural network computations are usually assumed to emerge from patterns of fast electrical activity. Challenging this view, we show that a male fly's decision to persist in mating hinges on a biochemical computation that enables processing over minutes to hours. Each neuron in a recurrent network contains slightly different internal molecular estimates of mating progress. Protein kinase A (PKA) activity contrasts this internal measurement with input from the other neurons to represent accumulated evidence that the goal of the network has been achieved. When consensus is reached, PKA pushes the network toward a large-scale and synchronized burst of calcium influx that we call an eruption. Eruptions transform continuous deliberation within the network into an all-or-nothing output, after which the male will no longer sacrifice his life to continue mating. Here, biochemical activity, invisible to most large-scale recording techniques, is the key computational currency directing behavior and motivational state.

Keywords: Drosophila; FLIM; PKA; cAMP; eruption; evidence accumulation; motivation; neural networks; sexual behavior; time.

Figure 1:. Four Crz neurons form a recurrent network that reports a 6-minute interval

A) Far left: Schematic summary of the CaMKII timer in the Crz neurons. Center left: Crz activity triggers a switch in behavioral state that permits termination in response to both opposing drives and the endpoint of the copulation. Center right and far right: Optogenetic inhibition of the Crz neurons with ACR1 beginning at various times into mating. Silencing the Crz neurons before 6 minutes into mating extends copulation duration from ~23 minutes, and often for longer than 1 hour (green), whereas silencing the neurons after 6 minutes has no effect on copulation duration. Individual mating durations are plotted as gray dots; the green curve demarcates the proportion classified as long matings at each time point (see Methods). N values and statistical analyses are listed in Supplementary Table 2. Unless otherwise noted, error bars throughout represent 68% credible intervals, selected to approximate SEM. B) Simultaneously measured CaMKII timers monitored with green-Camuiα (schematized on left) in two Crz neurons of the same male after optogenetic stimulation using ChR2-XXM (Scholz et al., 2017). Fluorescence lifetime is measured by pooling all pixels within an ROI and computing the empirical lifetime in each frame (see Methods). C) Measuring any individual CaMKII timer is informative about elapsed time, but cannot match the accuracy of the male’s behavior (“ability to estimate interval” refers to the mutual information between green-Camuiα measurements and a binary variable indicating the amount of time on the x axis has elapsed, see Methods and Figure S1). D) Pooling CaMKII activity across time (blue) and across neurons (green) permits a much more reliable estimate of elapsed time, especially if values are compared on the scale of ~10 seconds. E) Synaptic (Synapotagmin::GFP, Syt, green) and dendritic (Denmark, red) markers are closely intermingled in the Crz neurons. F) Optogenetic stimulation of individual Crz neurons (dark red traces) excites the other Crz neurons (light red traces). G) Probabilistic silencing (50%, using Coin-FLP(Bosch et al., 2015)) of each Crz neuron lengthens copulation in a proportion of males that suggests the requirement of only two active Crz neurons for network function (long matings whenever 3 or 4 neurons are silenced). H) Sustained optogenetic stimulation of even a single Crz neuron throughout mating (stochastically selected using a heat-inducible flippase (Nern et al., 2015); monitored by tdTomato fusion to Chr; red) is sufficient to recover a normal copulation duration when all Crz neurons are otherwise silenced by expression of Kir2.1 (monitored by EGFP fusion; cyan).

Figure 2:. The Crz neurons require cAMP signaling to reach a consensus.

A) A voltage-dependent signal is required at the end (consensus period) of the CaMKII timer to induce the switch in motivational state. B) Copulation duration of flies in which the Crz neurons are inhibited with ACR1 throughout mating except for a period of relief from inhibition. Providing windows of relief from inhibition after the CaMKII timers have expired shows that the Crz neurons require ~60–75 seconds of electrical activity to cause matings to terminate with normal, ~23 min durations. If relief is not provided, matings last much longer (>50 min). C) Proportion of matings categorized as long when inhibition of the Crz neurons was imposed throughout copulation except for two 50 second periods, spaced apart by a variable period of inhibition. D) Proportion of matings categorized as long after inhibition with a 30-second period of relief (x axis) or 90-second period of relief (y axis) across individual genotypes. A screen through 1,388 genetic manipulations of the Crz neurons (assayed as in Figure 2B) uncovered many hits in the cAMP signaling pathway. Colored dots indicate manipulations that are predicted to promote cAMP accumulation rut: rutabaga, an adenylyl cyclase; AC3: adenylyl cyclase 3; PDE4: a phosphodiesterase; KD: RNAi knockdown. Screen data available on request. Further characterization of hits in the screen is provided in Figure S2. E) Proportion of matings categorized as long after inhibition with variable periods of relief during manipulations of Gα_s activity in the Crz neurons (OE: overexpression). F) Proportion of matings categorized as long after inhibition with variable periods of relief during manipulations of cAMP levels in the Crz neurons. G) bPAC, a bacterial photoactivatable adenylyl cyclase, permits light-gated induction of cAMP signaling. H) Optogenetic induction of cAMP synthesis in the Crz neurons causes isolated male flies to ejaculate. I) bPAC-mediated cAMP synthesis is capable of prematurely inducing the switch in motivational state, as measured by termination in response to potentially lethal heat threats 5 minutes into mating. J) Proportion of matings categorized as long after inhibition with variable periods of relief with optogenetic synthesis of cAMP using bPAC in the Crz neurons.

Figure 3:. PKA translates cAMP into electrical activity and network output.

A) Schematic showing PKA activation by cAMP. B) Proportion of matings categorized as long after inhibition with variable periods of relief during knockdown of the PKA catalytic subunit PKA-C1 in the Crz neurons. C) Copulation duration of mating flies with and without inhibition of PKA activity using PKA-R* in the Crz neurons. D) Proportion of matings categorized as long after inhibition with variable periods of relief with expression of a constitutively active PKA catalytic subunit (PKA-mC*) in the Crz neurons. E) Calcium responses to cAMP synthesis in the Crz neurons measured ex vivo using GCaMP6s. Optogenetic induction of cAMP synthesis using bPAC results in a massive increase in intracellular calcium levels that requires PKA activity (each trace indicates one neuron from separate flies). F) Transmembrane calcium influx is required for the PKA/cAMP-mediated increase in intracellular calcium, as the GCaMP6s response is abolished by extracellular application of the Ca²⁺-channel blocker cadmium. G) Voltage dynamics of the Crz neurons in response to bPAC stimulation. cAMP signaling depolarizes the Crz neurons, as indicated by the voltage sensor ASAP2s. H) PKA-induced transmembrane calcium influx requires membrane depolarization, as it is blocked by expression of the leak K⁺ channel Kir2.1. I) Proportion of matings categorized as long after inhibition with variable periods of relief during knockdown of the β subunit of the voltage-gated calcium channel. J) Model illustrating how cAMP signaling drives consensus between the Crz neurons. Gα_s-signaling activates adenylyl cyclases, resulting in cAMP synthesis. cAMP increases PKA activity, which induces calcium influx through voltage-gated calcium channels (VGCCs). Intracellular calcium elevation drives the Crz neurons to signal to each other.

Figure 4:. cAMP accumulates within the Crz neurons to drive eruptions of activity

A-D) Ex vivo measurements of fluorescent cAMP pathway reporters following acute bPAC stimulation. Brief optogenetic induction (a 500 ms pulse) of cAMP synthesis using bPAC elevates cAMP levels (B) and PKA activity (C) without dramatically changing intracellular calcium (A) (inset: zoomed in traces from the larger panel). The resultant elevation of cAMP signaling lasts longer than intracellular calcium (D). In panel (D), each individual trace was normalized to its maximum (going from 0 at onset to 1 at its peak) and then these normalized traces were averaged together. E) Repeated stimulation of cAMP synthesis with bPAC every 100 seconds results in accumulation of intracellular cAMP. Residual cAMP and calcium are computed as the average value during the period 0.25 to 5 seconds preceding the next stimulation (i.e. 95–99.75 seconds after the most recent stimulation for 100 second intervals between each pulse). F-G) Accumulation of cAMP signaling eventually results in a massive increase in intracellular calcium levels that we term an eruption. An individual neuron is presented in (F), while (G) shows the dynamics of several neurons, each from a separate fly.

Figure 5:. Eruptions are synchronous across the Crz neuron network

A) Increases in intracellular cAMP in response to electrical stimulation, as indicated with caDDis_Green-Down. B,C) Simultaneous measurement of calcium activity of all Crz neurons during optogenetic induction of cAMP synthesis. Eruptions are synchronous across all Crz neurons (for details on the estimation of mutual information, see Figure S4). D-F) Top row: 500ms pulses of bPAC activation every 100 seconds results in an eruption approximately every fourth pulse on average (D). Knockdown of Gα_s within the Crz neurons (E), or of the active zone protein Unc13 (F), prevents eruptions. Bottom row: Prolonged bPAC activation (5 s) can induce eruption-like increases in calcium even when Unc13 and Gα_s are knocked-down. G) Likelihood of an individual 500 ms pulse of blue light eliciting an eruption (top) and the size of eruptions evoked by a 5 s pulse of blue light (bottom). Knockdown of Unc13 and Gα_s nearly abolish accumulation eruptions (top row), and reduce the magnitude of induced eruptions from sustained bPAC stimulation (bottom row). H) Model for a positive feedback loop between cAMP and electrical activity that concludes with an eruption.

Figure 6:. CaMKII delays the eruption by preventing PKA activation

A) Proportion of matings categorized as long after inhibition with variable periods of relief either 1 minute or 10 minutes into mating. B) Proportion of matings categorized as long after inhibition of the Crz neurons with 40 seconds of relief either 1 minute or 10 minutes into mating, supplemented by 40 seconds of stimulation of bPAC using blue light just before relief. C) Proportion of matings categorized as long after stimulation of bPAC in the presence of CaMKII-T287D. D) Measurement of cAMP levels with and without CaMKII-T287D in response to optogenetic stimulation of bPAC. E) Measurement of PKA activity with and without CaMKII-T287D in response to optogenetic stimulation of bPAC. F) Measurement of calcium levels with and without CaMKII-T287D in response to optogenetic stimulation of bPAC. G) Schematic showing how CaMKII delays the eruption by preventing network-activity-driven cAMP accumulation.

Figure 7:. cAMP signaling can track arbitrarily-patterned evidence over a range of timescales

A) The latency to terminate mating after an eruption was modeled as a Gaussian random variable. In this instance the Crz neurons were optogenetically inhibited for the first 10 minutes of mating, leading to a presumed eruption at 11 minutes, and a distribution of termination times centered on 29 minutes, with the posterior distributions of the fit parameters plotted in red and blue. Similar ~18 min delays were seen with various durations of inhibition. B) Flies expressing GtACR1 were exposed to 10 minutes of inhibition followed by cycling bouts of inhibition (pulse width) and permitted activity (window length), varying both parameters. C) Top row: Individual copulation durations of flies subjected to a fixed window length (indicated at top) and a variety of durations of pulsed inhibition. Longer inhibitory pulses widths always corresponded to a lengthened copulation duration, while increasing the relaxation window length for a fixed pulse width resulted in an earlier termination of mating. Bottom row: Inferred probability distribution for the cumulative electrical activity before an eruption in each condition by using the Gaussian model fit in Figure 7A. Longer copulation durations correspond to a later inferred eruption (see Methods). D) Left: For each window length, the cumulative activity required to trigger an eruption increased linearly with pulse width. Middle: Plotting the relationship between total time inhibited and total time active shows a fixed relationship across conditions, with a slope of ~¼ – one second of electrical activity negates four seconds of inhibition, regardless of the history of inhibition. Right: The constant of proportionality (~¼) between activity and inhibition is the same regardless of inhibition paradigm. E) Left: optogenetic inhibition was patterned by drawing sequential durations of inhibition and relieved inhibition from an exponential distribution. The mean duration of each pulse of inhibition was specified, as was the average fraction of the time the light was on, but otherwise each pulse was randomized. The randomized phases of inhibition were sustained throughout the mating. Middle: Using only the 1:4 rule derived in Figure 7D, we estimated how long a perfect integrator would need to reach a threshold equivalent to 75 seconds of unimpeded electrical activity (red line). When the fraction of the time the neurons are silenced exceeds 80% (i.e. there is at least 4 times as much inhibition as permitted activity), the integrator struggles to reach the threshold, though the increased variance with increasing pulse width means that for long pulse durations the system can still occasionally reach threshold. Right: Actual copulation duration data (circles) compared to the prediction of the integrator model (violin plots). The model predicts the distribution of copulation durations well, including the possibility of triggering an eruption even with the neurons silenced 85% of the time when the pulse width was long.