Behavioural choice emerges from nonlinear all-to-all interactions between drives

Under the right conditions any drive can overcome nearly any other, yet studies of behavioural selection predominantly focus on only one, or occasionally two behaviours. We present an experimental and computational framework that captures and explains the resolution of conflicts between several competing motivations. We characterize neurons that integrate information from all rival drives to generate an aggregate signal that urges male Drosophila to transition out of mating. Experimental investigation of these Drive Integrating Neurons (DINs) revealed time-varying, supralinear interactions among competing drives that stimulate the DINs and induce a change in behaviour. Extending these findings to model the interactions between all of an animal’s motivations led to the surprising prediction that, under many conditions, all-to-all interactions actually buffer the dominant drive against challengers. We experimentally validated this prediction, showing that weak drives for a variety of tertiary goals can have a profound stabilizing effect on the ongoing behaviour. These results emerge only if non-linear integration of other motivations occurs for each of an animal’s drives, suggesting the potential universality of this mechanism. Our findings emphasize the interconnectedness of motivational systems and the consequent importance of considering the full motivational state of an animal to understand its behaviour.


19
Animals often have multiple unmet needs, and attempting to satisfy one generally precludes 20 pursuing the others 1 . No one drive is strictly dominant; under the right conditions the pursuit of 21 nearly any goal may be suppressed by another 2 . At some level behaviour-specific drive states 22 must therefore affect the circuitry underlying many other behaviours 3 , and this information must 23 be integrated to arrive at a consensus. The ethologist Konrad Lorenz used the metaphor of a 24 "great parliament of instincts" to describe the behaviour of animals 2 , and the philosopher and 25 mathematician Bertrand Russell noted in his Nobel Prize acceptance speech that "If you wish to 26 know what men [sic] will do, you must know…the whole system of their desires with their 27 relative strengths" 4 . Nearly all studies on the interactions between competing motivations, in 28 contrast, focus on the resolution between just two drives in conflict. Here we establish an 29 experimental and computational framework for examining the many interactions between 30 simultaneous drive states that must be considered to understand naturalistic decision-making.

32
The mating duration of Drosophila melanogaster provides a clear and quantitative readout of the 33 interplay between competing drives: to switch behaviours the male must first terminate the 34 mating. If undisturbed, copulation will last ~23 minutes; if a dangerous situation arises, the male

46
Drive Integrating Neurons (DINs) control the decision to terminate mating

48
Constitutively silencing the DINs with tetanus toxin extends the average duration of mating from 49 ~23 minutes to ~1.5 hours 5 . To examine their moment-to-moment function during mating and in 50 response to threats, we conditionally silenced the DINs with GtACR1 (ACR1), a green-light 51 gated chloride channel. While tonic optogenetic silencing extended copulation duration to a 52 similar extent as tetanus toxin (Figure 1b), relaxing the inhibition at 30 minutes caused near-53 immediate termination of mating (Figure 1c). Inversely, turning on the light just before the 54 normal time of termination (at 20 minutes) most often caused matings to last well over an hour 55 (Figure 1c). These results show that electrical activity in the DINs is not required for tracking 56 time during copulation but instead causes termination after the appropriate time has passed.

57
Consistent with this interpretation, relaxing inhibition either shortly before the usual time of 58 termination (20 minutes) or at 5 minutes into the mating allowed copulations to terminate at the 59 appropriate time (Figure 1c). The temporal precision of these experiments shows that DIN 60 activity is only required around the time of mating termination, overturning the idea derived from

94
In line with their requirement only around the moment of natural termination, transient DIN 95 inhibition during the presentation of a threat caused the male to persist through severe 96 challenges that would otherwise truncate nearly all late-stage matings (Figure 1d and Video 1), 97 but did not extend copulation beyond its natural termination time (Figure 1e). The DINs are 98 therefore specifically required to make the decision to terminate mating, as we confirmed using 99 the heat-sensitive synaptic silencing tool UAS-Shibire ts 8 (Extended Data Figure 1), ruling out 100 tool-specific artifacts (e.g. due to changes in the chloride equilibrium potential 9 ). We conclude 101 that DIN activity ends copulation in response to two types of triggering stimuli: (i) competing 102 drives (e.g. survival in the case of heat threats); and (ii) the fulfillment of all mating goals at ~23 103 minutes. At the level of DIN activity, there appears to be little, if any, difference between these 104 two classes of demotivating conditions.

106
Using the red-light gated cation channel CsChrimson (Chr) 10 we found that acute stimulation

166
In the simplest such model, the instantaneous probability of responding to a stimulus behaves 167 like a linear dynamical system with time constant . This enabled us to fit the cumulative 168 distribution of actual times of termination to the equation (where ! is the strength of the demotivating input) to estimate the parameters of the model 171 (Extended Data Figure 3a, see Methods for more information). We reduced the strength of 172 optogenetic activation by using green light, which penetrates tissue less effectively than red 11 , 173 allowing us to stimulate the DINs for longer durations without immediately ending the mating. A 174 shorter caps the peak termination probability for long threats of fixed intensity, whereas 175 extending integration to longer timescales leads to an increased output and, therefore, 176 increased termination probability.

178
To assess the overall fit of the data we plotted cumulative termination probabilities, which show 179 that the data is well fit by a linear integration process in time, with ≈ 1-2 seconds at 10

185
The fit is orders of magnitude better when temporal integration is included than if the DINs are 186 assumed not to integrate over time (Figure 2e). This analysis provides quantitative estimates 187 for the expanding time constant as mating progresses, arguing that the response to competing 188 drives is increased as the mating progresses by lengthening the integration time of the DINs.

189
We emphasize that this model is descriptive: it does not claim that the parameters and ! truly 190 exist in some physical form. Instead, it provides a means to describe and analyze how the DINs 191 integrate information over time and allows us to assess the demotivating impact of a memory-192 like component within this decision-making circuitry.

194
A temporal window of integration should potentiate responses not just to sustained challenges, 195 but also to discrete inputs separated in time. We therefore stimulated the DINs with paired 500 196 ms excitatory pulses separated by 0-to-30 seconds at 10 or 15 minutes into mating. When the 197 two DIN pulses were supplied in near-immediate succession, we found an augmentation of the 198 second pulse at both 10 and 15 minutes (Extended Data Figure 4b). When the pulses were 199 spaced out, augmentation was still evident with at longer inter-pulse intervals later in mating, 200 closely matching the effects seen with sustained stimulation.

202
Motivating inputs limit the ability of competing drives to activate the DINs

204
We used this quantitative analysis to ask whether inputs that motivate sustained copulation act  (Figures 3a,b). Sustained

280
Temporal integration could be implemented in many ways, but because direct optogenetic 281 stimulation of the neurons was integrated across time, (Figures 2c-e), it seemed unlikely that 282 the effect arises from changing dendritic responses to synaptic input. We therefore focused on  Figure 5m), but also resulted in a sustained increase in axonal calcium that persisted for 289 tens of seconds, much longer than the off-kinetics of the reporter itself 15 . As predicted by a 290 model in which residual calcium mediates the augmented behavioural response, sustained 291 elevations of calcium were enhanced following a second stimulating pulse (Figures 3e,f), with 292 no discernible effect on peak calcium (Figure 3g). Application of dopamine to the bath nearly 293 abolished the sustained elevations in calcium after optogenetic excitation (Figures 3e,f)

301
The DINs synergistically pool demotivating inputs across modalities

303
Every threatening, damaging, or otherwise demotivating stimulus to which we have subjected a 304 mating pair (short of forcible separation) requires the activity of the male's DINs to elicit a 305 termination response 5 . We sought optogenetically-tractable behaviours that could oppose 306 copulation to test whether the principles derived from direct DIN stimulation apply to other 307 drives. Stimulating neurons that drive grooming behaviour terminates mating with increasing 308 efficacy as the mating progresses (Figure 4a), and required DIN activity to do so (Figure 4a).

309
Grooming itself, whether induced by optogenetic stimulation (Video 4) or application of baking 310 flour (Video 5), was suppressed during mating, but was initiated rapidly upon termination,

311
showing that this paradigm resulted in a genuine competition between the two behaviours.

312
Grooming behaviour showed the same characteristics of temporal integration as direct DIN 313 stimulation (Extended Data Figures 4c-g, Supplementary Discussion 1). Since demotivating 314 stimuli of each modality converge at the DINs, and since paired DIN stimulations produce a 315 synergistic response greater than the sum of their independent probabilities, we predicted that 316 multimodal competing inputs would combine to generate a stronger termination response than 317 when delivered alone, or even than their independent sum, when delivered together. Confirming

339
We next performed a series of experiments pairing direct DIN stimulation with heat (Figure 4c).

340
When heat and DIN stimulation were paired before the Crz-neuron mediated switch in 341 motivation at ~6 minutes into mating, we saw no contribution of a strong heat threat to 342 termination probability (Figure 4c, also see Supplementary Note 2). This rules out any 343 enhancing effect of temperature on CsChrimson activation itself 16 , and corroborates our finding 344 that access to the DINs by real-world demotivating stimuli is blocked before the Crz switch is

354
To extend these findings, we sought other impulses that could compete with copulation. We 355 screened a collection of split-Gal4 lines that label small sets of neurons with cell bodies in the 356 brain and that send projections to the VNS 7 for lines that could oppose the motivation to 357 copulate (Extended Data Figure 4i). The response to prolonged stimulation of the most 358 effective of these, AG Desc (for Abdomingal Ganglion-projecting Descending neurons) (Extended 359 Data Figure 4j,k), showed augmentation over a time course that resembled that seen with the 360 grooming neurons (Extended Data Figure 4l). As with grooming neuron stimulation, pairing 361 excitation of the AG Desc with heat threats resulted in a greater termination probability than could 362 be explained by the two stimuli acting independently (Figure 4d).

364
These results point to drive integration at the DINs or elsewhere in the nervous system. To test 365 integration by the DINs, themselves, we combined heat threats with brief grooming stimulation 366 and silenced the DINs selectively during the stimulation of the grooming neurons (Figure 4e).

367
This returned the termination probability to that of the heat threat alone (Figure 4e), arguing, 368 together with the above results, that the integration of competing information occurs at the DINs.

370
High-order interactions between drives can stabilize or destabilize the ongoing behaviour

372
The data presented above suggest three novel principles of motivational control over behavioral 373 selection: i) synergistic effects of multimodal competing drive inputs on the ongoing behavior; ii) 374 long-timescale integration of diverse inputs at behavior-specific demotivating neurons; and iii) 375 that motivational cues prevent or limit integration of competing drive inputs. In this section we 376 explore the implications of these principles assuming they hold across many or all behaviours.

378
We generated an integration-based mathematical model for high-order (i.e. supralinear)

379
interactions between multiple drives. Drives are represented as evolving variables in a 380 dynamical system using the principles described above, which we assessed both numerically

451
To experimentally test the surprising prediction of stabilizing effects of weak tertiary drives 452 (Figure 5d), we delivered heat threats during mating while also optogenetically activating the  Figure 10). Remarkably, the combined terminating impact of 457 heat and the tertiary drive was usually lower than the heat threat alone (e.g. Figure 5e, 500ms 458 pulses). Though we emphasize that we do not know which drives are promoted when we 459 stimulate most of these lines, the results are strikingly consistent with the prediction from the 460 model, showing that a weak tertiary drive can dramatically decrease the effectiveness of a 461 strong challenger (the heat threat). By increasing the duration of optogenetic stimulation, we 462 found that if-and only if-increased stimulation turned these tertiary drives into strong 463 challengers (i.e. they often overcame mating drive even when presented in isolation), they then 464 synergized with heat to cause termination rates higher than would be expected from the

475
The great parliament of instincts

477
Our results argue that Lorenz's metaphor of a parliament of instincts may be useful beyond the 478 immediate mental imagery it conjures: "it is a more or less complete system of interactions 479 between many independent variables" in which "all imaginable interactions can take place

486
Our proposed mechanistic instantiation of the parliamentary model for behavioural selection 487 predicts that demotivating neurons like the DINs will be found to regulate many behaviours. The 488 clearest analogs that we see in the literature are the parabrachial CGRP neurons of the 489 mammalian hypothalamus. These neurons prevent feeding when activated, are stimulated by a 490 wide variety of aversive cues, and are themselves suppressed by the AgRP neurons that 491 motivate feeding behaviour [17][18][19] . Silencing the CGRP neurons leads to extended bouts of 492 feeding 19 arguing that they are required to demotivate feeding not just in response to competing 493 drives, but also with the onset of satiety. Hunger is the most intensively studied mammalian 494 motivation, and we expect that the ongoing circuit-level investigations into other behaviours will 495 uncover demotivating nodes with converging inputs from many opposing drives.

497
Temporal integration by demotivating neurons

499
Most adjustments to circuit functions over time and with experience have been found, or 500 assumed, to result from changing synaptic weights. Here we show that a demotivating node 501 operates over long motivational and decision-making timescales through a different mechanism 502 for altering the response to fixed input: changing the time constant of integration. This 503 mechanism has several theoretical advantages. For example, the representations of stimuli 504 shorter than the timescale of integration are preserved, avoiding a potentiation of the response 505 to noise and acting as a tunable low-pass filter.
While we cannot yet provide a detailed mechanistic description of the changing time constant of 508 integration in the DINs, we find it useful to think about it in terms of the well-known phenomenon 509 of synaptic augmentation 14 . Though not mechanistically understood itself, synaptic 510 augmentation is thought to emerge from lingering calcium after an initial stimulus, generating a 511 seconds-long period of increased transmitter release probability that decays as localized 512 calcium is buffered or cleared. We also observe lingering calcium in the DINs, and find that the 513 augmentation persists through electrical silencing, indicating that the memory-like trace is stored 514 and adjusted biochemically. Synaptic augmentation lasts up to tens of seconds 20 , with a time 515 constant that is independent of the strength of the initiating stimulus, also similar to the effects 516 we observe here. In the DINs, augmentation is tuned by motivational inputs like dopamine to 517 alter the impact of contemporaneous or long-lasting challenges as the male progresses through 518 the mating. Targeted, functional genetic screening of the DINs will likely reveal the mechanisms 519 that adjust this signal and implement its effects, information that may bring us to the verge of a 520 thorough molecular explanation of motivation in this system.