Part 1:A Dopamine Gradient Controls Access To Distributed Working Memory in The Large-scale Monkey Cortex
Mar 19, 2022
Contact: Audrey Hu audrey.hu@wecistanche.com
Sean Froudist-Walsh,1 Daniel P. Bliss,1 Xingyu Ding,1 Lucija Rapan,2 Meiqi Niu,2 Kenneth Knoblauch,3,4 Karl Zilles,2,8 Henry Kennedy,3,4,5,7 Nicola Palomero-Gallagher,2,6,7 and Xiao-Jing Wang1,7,9,*
1Center for Neural Science, New York University, New York, NY 10003, USA
2 Research Centre Julich, INM-1, Julich, Germany
3 INSERM U846, Stem Cell & Brain Research Institute, 69500 Bron, France
4 Universite´ de Lyon, Universite´ Lyon I, 69003 Lyon, France
5 Institute of Neuroscience, State Key Laboratory of Neuroscience, Chinese Academy of Sciences (CAS), Key Laboratory of Primate Neurobiology CAS, Shanghai, China
SUMMARY
Dopamine is required for working memory, but how it modulates the large-scale cortex is unknown. Here, we report that dopamine receptor density per neuron, measured by autoradiography, displays a macroscopic gradient along with the macaque cortical hierarchy. This gradient is incorporated in a connectome-based large-scale cortex model endowed with multiple neuron types. The model captures an inverted U-shaped dependence of working memory on dopamine and spatial patterns of persistent activity observed in over 90 experimental studies. Moreover, we show that dopamine is crucial for filtering out irrelevant stimuli by enhancing inhibition from dendrite-targeting interneurons. Our model revealed that an activity-silent memory trace can be realized by the facilitation of inter-areal connections and that adjusting cortical dopamine induces a switch from this internal memory state to distributed persistent activity. Our work represents a cross-level understanding from molecules and cell types to recurrent circuit dynamics underlying a core cognitive function distributed across the primate cortex.
INTRODUCTION
Our ability to think through difficult problems without distraction is a hallmark of cognition. When faced with a constant stream of information, we must keep certain information in mind and protect it from distraction. For instance, when at the supermarket looking for your favorite butter, it is important to keep in mind its distinctive golden packaging and not be distracted by the many other dairy products. This brain function is called working memory. Working memory often engages persistent neural activity that is specific to the information that must be remembered. This mnemonic activity is sustained internally across multiple cortical and subcortical areas in the absence of external stimulation (Funahashi et al., 1989; Fuster and Alexander, 1971; Guo et al., 2017; Leavitt et al., 2017; Mejias and Wang, 2021; Men-doza-Halliday et al., 2014; Murray et al., 2017; Romo et al., 1999; Romo and Salinas, 2003; Vergara et al., 2016; Wang, 2001; Zhang et al., 2019).
Working memory and the prefrontal cortex are under the influence of monoaminergic modulation (Goldman-Rakic, 1995; Robbins and Arnsten, 2009). In fact, depletion of dopamine from the prefrontal cortex and complete ablation of the prefrontal cortex cause similar working memory deficits (Brozoski et al., 1979). Dopamine modulates cortical activity through its receptors. D1 receptors are the most densely expressed dopamine receptor type in the cortex. Prefrontal neuron activity during working memory depends on precise levels of activation of D1 receptors, with too little or too much D1 stimulation disrupting delay period activity (Vijayraghavan et al., 2007; Wang et al., 2019). However, the density of D1 receptors is known only for relatively small sections of the monkey cortex (Goldman-Rakic et al., 1990; Impieri et al., 2019; Lidow et al., 1991; Niu et al., 2020; Richfield et al., 1989). Because of the shortage of areas analyzed across studies, it is not clear whether the variation in D1 receptor densities across cortical areas represents random heterogeneity or a systematic gradient of cortical dopamine modulation.
Dopamine receptors are also expressed differently across different types of inhibitory neurons (Mueller et al., 2018, 2020). Distinct inhibitory cell types primarily focus their inhibition on the dendrites or somata of pyramidal cells or on other inhibitory neurons (Jiang et al., 2015; Tremblay et al., 2016). Through its differing effects on distinct interneurons, dopamine decreases inhibition to the somata of pyramidal cells and increases inhibition to the dendrites (Gao et al., 2003). An early theoretical study proposed that inhibition targeted more strongly toward the dendrites and away from the somata of pyramidal cells could increase the resistance of working memory to distraction (Wang et al., 2004a). The functional significance of dopamine’s differential effects on distinct inhibitory neuron types has not yet been investigated.
In this work, we tackled two open questions. First, how does dopamine modulate distributed working memory across a multi-regional large-scale cortical system? Second, in light of the emphasis on cell types in modern cortical physiology, does dopamine contribute to robust working memory against distractors by virtue of differential effects on different neuron classes? To address these questions, we performed quantitative mapping of dopamine D1 receptor densities across 109 cortical areas using in vitro autoradiography and constructed a large-scale computational model of the macaque cortex that is capable of performing working memory tasks. The model is built using retrograde tract-tracing connectivity data and incorporates gradients of D1 receptors and excitatory synapses. Moreover, to our knowledge, this is the first large-scale cortex model endowed with three subtypes of inhibitory neurons. Our results suggest that firing of dopamine neurons can engage distractor-resistant, stimulus-selective, sustained activity across multiple brain regions in response to behaviorally relevant stimuli. Furthermore, we extend, from a local area to the multi-regional cortex, an activity-silent mechanism that has been proposed for certain forms of short-term memory trace without persistent activity (Mongillo et al., 2008; Rose et al., 2016; Wolff et al., 2017). We found that this scenario relies principally on short-term facilitation of inter-areal connections but fails to resist distractors. Enhanced dopamine modulation can convert an internal memory trace to an active persistent activity state needed to filter out distractors. Therefore, our findings contribute to resolving the current debate about the two contrasting scenarios that contribute to working memory (Constantinidis et al., 2018; Lundqvist et al., 2018; Wa- Tanabe and Funahashi, 2014) and under what conditions each mechanism is implemented (Barbosa et al., 2020; Masse et al., 2019; Trbutschek et al., 2019).
RESULTS
A hierarchical gradient of dopamine D1 receptors per neuron across the monkey cortex
We first analyzed D1 and D2 receptor distribution patterns throughout the macaque brain using in vitro receptor autoradiography (Figure S1). Autoradiography enables quantification of endogenous receptors in the cell membrane through the use of radioactive ligands (Niu et al., 2020; Palomero-Gallagher and Zilles, 2018; Rapan et al., 2021). The highest densities (in fmol/ mg protein) of both receptor types were found in the basal ganglia, with the caudate nucleus (D1, 298±28; D2, 188± 30) and putamen (D1, 273±40; D2, 203±37) presenting considerably higher values than the internal (D1, 97±34; D2, 22± 12) or external (D1, 55±16; D2, 30±11) subdivisions of the globus pallidus. Raw cortical D1 receptor densities ranged from 49± 13 fmol/mg protein in area 4a of the primary motor cortex to 101±35 fmol/mg protein in orbitofrontal area 11l (Figure 1A).
The density of the D2 receptor in the cortex is so low that it is not detectable with the method used here.
To compare the gradient of D1 receptors with other known gradients of anatomical organization in the monkey cortex, we carefully mapped the receptor data (Figure 1A) as well as data on neuronal density (Figure 1B; Collins et al., 2010) and spine count (Figure 1C; Elston, 2007) onto the Yerkes19 common cortical template, to which anatomical tract-tracing data (Figure 1D, i) has been mapped previously (Donahue et al., 2016). Here we include retrograde tracing data from 40 regions, quantified using the same protocol as in previous publications (Markov et al., 2014b). This expands the number of injected cortical areas by 33%, with connections to areas 1, 3, V6, F4, F3, 25, 32, 9, 45A, and OPRO (orbital proiso cortex) now included in the database (downloadable from core-nets.org). We estimated the cortical hierarchy using laminar connectivity data (Figure 1D, ii; STAR Methods; Markov et al., 2014a), expanding previous descriptions of the cortical hierarchy based on fewer regions (Markov et al., 2014a; Mejias et al., 2016). A one-dimensional hierarchy is probably an oversimplification of the cortical connectivity structure. Because we have connectivity data for two distinct sensory modalities, we also calculated a circular embedding of the connectivity data, with radial distance from the edge representing the hierarchical position and angular distance between points representing the inverse of their connectivity strength (Chaudhuri et al., 2015). In this circular representation, separate visual and somatosensory hierarchies can clearly be appreciated, with association regions falling at angles off the main sensory hierarchy axes (Figure 1E).
To facilitate functional interpretation, we divided D1 receptor density by neuron density (Collins et al., 2010) to allow estimation of the degree to which dopamine modulates individual neurons across the cortex. D1 receptor density per neuron peaked in the parietal and frontal cortex and was relatively low in the early sensory cortex (Figure 1F). There was a strong positive correlation between D1 receptor density per neuron and the cortical hierarchy (Figure 1G; r = 0.81). Because of spatial auto-correlation between cortical features (i.e., nearby parts of the cortex tend to have similar anatomy), it is possible to detect spurious correlations between distinct features of brain anatomy. To account for this, we generated 10,000 surrogate maps with similar spatial autocorrelation to the hierarchy map (Burt et al., 2020). None of these surrogate maps were as strongly correlated with the D1 receptor density map as the hierarchy, giving a p-value of less than 0.0001 for the D1 receptor-hierarchy correlation. There was no significant relationship between D1 receptor expression and whether a cortical area had a granular layer IV (Wilcoxon rank-sum Z = 0.39, p = 0.70) or to the degree of Dexternopyramidalization (Kruskal- Wallis c2 = 1.47, p = 0.48; Goulas et al., 2018; Sanides, 1962; Figure S2). This pattern of receptor expression suggests that dopamine principally modulates areas contributing to higher cognitive processing.
We then built a large-scale model of the macaque cortex. We placed the local circuit in each of the 40 cortical areas (Figure 2A, right). Properties of these local circuits varied across areas in the form of macroscopic gradients (Wang, 2020) of long-distance connectivity (set by tracing data), the strength of excitation (set by the spine count), and modulation by D1 receptors (set by the receptor autoradiography data). We defined the connections between areas using the quantitative retrograde tract-tracing data. In the model, inter-areal connections are excitatory and target the dendrites of pyramidal cells (Petreanu et al., 2009). Inter-areal excitatory connections also target calretinin (CR)/vasoactive intestinal peptide (VIP) cells to a greater degree than parvalbumin (PV) or calbindin (CB)/somatostatin (SST) cells (Lee et al., 2013; Wall et al., 2016). The frontal eye fields (FEF) have an unusually high density of CR (here CR/VIP) cells (Pouget et al., 2009). To account for this, we increased the proportion of inter-areal input to CR/VIP cells in FEF and reduced the strength of input to PV and CB/SST cells.
An inverted U relationship between cortical D1 receptor stimulation and distributed working memory activity
We simulated the large-scale cortical model during the performance of a working memory task (Figure 2C) with different levels of cortical dopamine availability. In simulations, stimulus-selective Inter-areal connectivity determines the distributed working memory activity pattern. We next compared the pattern of delay period activity in the model with delay period activity observed in over 90 electro- physiology studies (Leavitt et al., 2017). We chose model parameters that would produce persistent activity in the prefrontal cortex, but we did not fit the model to the experimental data. Of the 19 cortical areas in which such activity has been assessed during the delay period in at least three experimental studies, 18 were in agreement between the simulation and experimental results (c2 = 15:03; p = 0:0001 Figure 3A). Overall, the experimentally observed persistent activity from numerous studies is reproduced, validating the model. This allows us to inspect the anatomical properties that underlie the distributed activity pattern and gain insight into the brain mechanisms that may pro-duce it.
We repeated model simulations after shuffling the anatomical data. The delay period activity patterns for 30,000 simulations based on the shuffled anatomy were compared with the pattern observed experimentally. Ten thousand simulations were run using shuffled inter-areal connections, shuffled D1 receptor expression, and shuffled dendritic spine expression separately. The overlap between the experimental persistent activity pattern and the model persistent activity pattern was strongly dependent on the inter-areal connections (p = 0.0004) but not on the pattern of D1 receptors (p = 0.71) or dendritic spine count (p = 0.46) (Figure 3B). This analysis suggests that the edges between nodes in the network (i.e., the inter-areal connections) are important for defining the spatial pattern of delay period activity. Next, we asked how the nodes themselves (i.e., individual cortical areas) contribute differentially to distributed working memory.
Working memory deficits are most severe following lesions to prefrontal areas with high D1 receptor density
We next quantified the degree to which focal lesions to individual areas in the model disrupted persistent activity during the working memory task (without distractors). The effect depended on the lesioned area and the level of cortical dopamine (Figure 3C). Lesions to prefrontal and posterior parietal areas caused the greatest reductions in delay period firing rates (Figure 3D, E). Lesions to frontal areas caused a significantly greater reduction in delay period firing rates than lesions to parietal areas (Mann- Whitney U = 46.0, p = 0.027). We tested the effects of progressively larger lesions on the frontal and parietal cortex. To increase the size of the lesions, for each lobe we first lesioned the area that caused the biggest drop in delay activity when lesioned individually and then additionally lesioned the area that caused the second-biggest drop and so on (frontal lesion 1: 46d, lesion 2: 46d+8B, lesion 3: 46d+8B+8 m, etc.; parietal lesion 1: LIP, lesion 2: LIP+7m, lesion 3: LIP+7 m+7B., etc.). When lesioning two frontal regions, the mnemonic delay period activity was completely destroyed throughout the cortex, so the network was no longer able to perform the task. In contrast, progressively larger lesions of the parietal cortex caused only a gradual decrease in frontoparietal delay activity, and even when the entire parietal cortex was removed (10 areas), sufficient residual mnemonic delay period activity remained to allow the cue stimulus to be decoded (Figure 3F).
We subsequently addressed the ability of the model to maintain cue-specific delay period activity in the presence of distractors following precise lesioning of each cortical area. We analyzed trials across all levels of cortical dopamine availability. Lesions to three prefrontal areas (8m, 46d, and 8B), but not other areas, caused complete disruption of distractor-resistant working memory activity in all trials. Lesions to many other areas caused a complete reduction of distractor-resistant working memory activity for some trials (corresponding to a particular dopamine range) but not others. The seven lesions causing the greatest disruption of working memory performance were in the frontal cortex (six prefrontal areas and premotor area F7; Figure 3G). The reduction in performance was significantly greater for lesions to frontal cortical areas than parietal areas (Mann- Whitney U = 48.5, p = 0.032). Our simulations thus suggest that (1) lesions to the prefrontal and posterior parietal cortex can cause a significant disruption of delay period activity, (2) frontal lesions have a greater effect on behavior than parietal lesions, and (3) smaller lesions, particularly to the prefrontal cortex, can significantly disrupt performance on more difficult working memory tasks, such as those with distractors. In contrast, larger lesions are required to disrupt performance on simple working memory tasks.
