1. Neuroscience
Download icon

Disruption of Nrxn1α within excitatory forebrain circuits drives value-based dysfunction

  1. Opeyemi O Alabi
  2. M Felicia Davatolhagh
  3. Mara Robinson
  4. Michael P Fortunato
  5. Luigim Vargas Cifuentes
  6. Joseph W Kable
  7. Marc Vincent Fuccillo  Is a corresponding author
  1. Department of Neuroscience, United States
  2. Neuroscience Graduate Group, Perelman School of Medicine, United States
  3. Department of Psychology, University of Pennsylvania, United States
Research Article
  • Cited 0
  • Views 501
  • Annotations
Cite this article as: eLife 2020;9:e54838 doi: 10.7554/eLife.54838

Abstract

Goal-directed behaviors are essential for normal function and significantly impaired in neuropsychiatric disorders. Despite extensive associations between genetic mutations and these disorders, the molecular contributions to goal-directed dysfunction remain unclear. We examined mice with constitutive and brain region-specific mutations in Neurexin1α, a neuropsychiatric disease-associated synaptic molecule, in value-based choice paradigms. We found Neurexin1α knockouts exhibited reduced selection of beneficial outcomes and impaired avoidance of costlier options. Reinforcement modeling suggested that this was driven by deficits in updating and representation of value. Disruption of Neurexin1α within telencephalic excitatory projection neurons, but not thalamic neurons, recapitulated choice abnormalities of global Neurexin1α knockouts. Furthermore, this selective forebrain excitatory knockout of Neurexin1α perturbed value-modulated neural signals within striatum, a central node in feedback-based reinforcement learning. By relating deficits in value-based decision-making to region-specific Nrxn1α disruption and changes in value-modulated neural activity, we reveal potential neural substrates for the pathophysiology of neuropsychiatric disease-associated cognitive dysfunction.

Introduction

Goal-directed behaviors are a critical aspect of animal fitness. Their implementation engages widespread neural circuits, including cortico-striatal-thalamic loops and midbrain dopaminergic populations. Cortical regions including orbital frontal (OFC), medial prefrontal (mPFC), and anterior cingulate (ACC) represent aspects of reward value and history (Bari et al., 2019; Bartra et al., 2013; Euston et al., 2012; Noonan et al., 2011; Padoa-Schioppa and Conen, 2017; Rushworth et al., 2011; Rushworth et al., 2012). Primary sensory cortices and midline thalamic nuclei represent reward-associated environmental signals (Parker et al., 2019; Znamenskiy and Zador, 2013) while motor thalamic nuclei ensure smooth performance of actions (Díaz-Hernández et al., 2018). Furthermore, flexible adaptation of value signals is supported by error-monitoring signals within ACC and basolateral amygdala, as well as reward prediction errors encoded by striatal-targeting midbrain dopaminergic neurons (McGuire et al., 2014; Schultz et al., 1997; Ullsperger et al., 2014; Yacubian et al., 2006). The dorsal striatum via integration of these diverse projections can simultaneously mediate action selection, motor performance, and reinforcement learning (Balleine et al., 2007; Cox and Witten, 2019; Lee et al., 2015; Vo et al., 2014).

Deficits in goal-directed decision making, and specifically in how reward shapes selection of actions, are a core endophenotype shared across neuropsychiatric disorders, including schizophrenia, autism spectrum disorders (ASD), obsessive-compulsive disorder, and Tourette syndrome (Barch and Dowd, 2010; Corbett et al., 2009; Dichter et al., 2012; Dowd et al., 2016; Gillan and Robbins, 2014; Griffiths et al., 2014; Hill, 2004; Maia and Frank, 2011; Solomon et al., 2015). In schizophrenia, impairments in action–outcome learning (Gold et al., 2015; Morris et al., 2015) may reflect perturbations to reinforcement learning error signals or the manner in which they are integrated to impact choice (Hernaus et al., 2019; Hernaus et al., 2018). Recent studies have also revealed reinforcement learning deficits in ASD patients (Hill, 2004; Solomon et al., 2015), with impaired choice accuracy driven by reduced win–stay choice patterns (Solomon et al., 2015).

Genetic association studies for neuropsychiatric disease have converged on synapses as key sites of disease pathophysiology (DDD Study et al., 2014; Willsey et al., 2013; Willsey and State, 2015). Neurexin1α (Nrxn1α) is an evolutionarily conserved synaptic adhesion molecule, for which rare de novo and inherited copy number variants confer significant risk for ASDs, schizophrenia, Tourette syndrome, and obsessive-compulsive disorder (Ching et al., 2010; Duong et al., 2012; Huang et al., 2017; Kirov et al., 2009; Lowther et al., 2017; Rujescu et al., 2009). The Neurexin family of proteins functions as a presynaptic hub for transynaptic binding of numerous postsynaptic partners at both excitatory and inhibitory synapses (Missler et al., 2003; Südhof, 2017). Consistent with their expression prior to synaptogenesis (Harkin et al., 2017; Puschel and Betz, 1995), Neurexins have been implicated in the initial specification and long-term integrity of synapses (Anderson et al., 2015; Aoto et al., 2013; Chubykin et al., 2007; Krueger et al., 2012; Soler-Llavina et al., 2011; Südhof, 2017; Varoqueaux et al., 2006). While Nrxn1α transcripts are broadly expressed throughout the brain, they are particularly enriched in cortico-striatal-thalamic loops proposed to govern motor control, action selection, and reinforcement learning (Fuccillo et al., 2015; Ullrich et al., 1995).

Behavioral abnormalities in Nrxn1α knockout animals include reduced nest building and social memory, increased aggression and grooming, enhanced rotarod learning, and male-specific reductions in operant responding under increasing variable interval responding schedules (Dachtler et al., 2015; Esclassan et al., 2015; Etherton et al., 2009; Grayton et al., 2013). Despite this broad dysfunction, the underlying mechanistic contributions of Nrxn1α to disease-relevant behaviors remain unclear, owing to our poor understanding of the specific computational algorithms and neural circuit implementations for the behavioral functions interrogated by these standard assays.

In this paper, we uncover widespread alterations in reward processing in Nrxn1α knockout mice, manifest as inefficient choice and altered control of task engagement. These deficits were observed across a range of value comparisons and feedback rates, suggestive of trait-like decision-making abnormalities. Modeling of choice patterns suggests these deficits are driven by impaired learning and representation of choice values. To reveal causal circuits for this reward processing defect, we performed brain region-specific deletion of Nrxn1α. We found that Nrxn1α disruption in excitatory telencephalic neurons, but not thalamic neurons recapitulated the choice and reward processing abnormalities of brain-wide Nrxn1α knockouts. Furthermore, telencephalic projection neuron-specific Nrxn1α disruption produced dysregulation of value-associated circuit activity prior to choice in direct pathway neurons of the dorsal striatum. Together, this work represents an important step in characterizing the genetic contributions to circuit dysfunction for a core neuropsychiatric disease-relevant behavior – how animals choose actions according to cost and benefit.

Results

Neurexin1α KOs have blunted responses to relative reward outcomes

We found that Nrxn1α knockout (KO) mice could perform basic light-guided operant responding with consistent task engagement (Figure 1—figure supplement 1A–C). Next, we specifically tested how Nrxn1α mutant mice use value information to guide future choice via a feedback-based paradigm (Figure 1A). Briefly, mice self-initiated consecutive two alternative forced-choice trials where each alternative was associated with contrasting reward volumes (12 μL versus either 0 μL, 2 μL, 6 μL, or 8 μL). To explore whether value comparisons were further influenced by reward scarcity, we tested four relative reward ratios in both high (Prew = 0.75) and low (Prew = 0.4) feedback regimes. Alternation of reward contingencies was used (triggered by 80% bias toward the larger reward in a moving 10-trial block) to maintain outcome sensitivity over hundreds of trials (Figure 1A; see Alabi et al., 2019 and Materials and methods for further details). Performance in this task was significantly altered by the relative magnitude of rewarded outcomes for both wild-type and KO animals with larger reward contrasts driving more biased choice patterns (Figure 1B). Nonetheless, we observed a global decrease in session performance across relative reward contrasts in Nrxn1α KO mice as compared to wild type (Figure 1B), without genotypic differences in total reward consumed or task engagement (Figure 2—figure supplement 1B and C).

Figure 1 with 1 supplement see all
Neurexin1α disruption leads to deficits in value-based selection of actions.

(A) Schematic of trial structure wherein mice perform repeated self-initiated trials with contrasting reward volumes associated with each port. Animals were tested at four relative reward ratios across high (Prew = 0.75) and low (Prew = 0.4) reinforcement rates. See Materials and methods for details. (B) Both probability of reinforcement and volume contrast modulate the probability at which mice select the large reward option. Nrxn1α KOs (blue, n = 10) select the high benefit alternative at a lower rate than their WT littermates (gray, n = 11) across reward environments (three-way RM ANOVA). (C and D) For both WT and KO animals, the relative magnitude of rewarded outcome has a significant effect on the stay-probability for that alternative. (E) The relative reward-stay (RRS), which quantifies the relative tendency of animals to repeat choices after specific outcomes, was sensitive to relative magnitude of rewards but not reward probability. In comparison to WT littermates, Nrxn1α KOs less dynamically alter their choice behavior after large reward outcomes than small reward outcomes (three-way RM ANOVA). (F and G) The RRS is a significant predictor of session performance for both WT and KO mice at both rates of reinforcement. Note RRS is a better predictor of task performance at high reinforcement rates, reflecting the preponderance of unrewarded outcomes in low reinforcement conditions. All data represented as mean ± SEM.

Performance could be altered by changes in: (1) how feedback is integrated over time; (2) sensitivity to outcome feedback; and (3) flexibility to changing contingencies (Alabi et al., 2019). To assess whether Nrxn1α KOs show altered influence of reward history on current choice, we employed logistic regression models to estimate the relative effects of choice and outcome (five preceding trials) on current choice (Lau and Glimcher, 2005; Parker et al., 2016; Tai et al., 2012). We found that wild-type mice and Nrxn1α KOs heavily discount all but the immediately preceding trial (t−1) (Figure 1—figure supplement 1E–H), suggesting a significant portion of choice variability can be accounted for by analyzing influences of the t−1 trial. We therefore calculated the relative reward-stay (RRS), a measure of the relative reinforcing properties of large versus small rewarded t−1 outcomes (previously relative action value in Alabi et al., 2019). We noted smaller gaps between large reward-stay and small reward-stay behavior in Nrxn1α KOs as compared with wild types (Figure 1C and D), leading to smaller RRS values across reward contrasts and feedback environment (Figure 1E). The significant correlation between RRS and performance across genotypes highlights the importance of outcome sensitivity on task performance (Figure 1F and G).

As deficits in behavioral adaptability have been observed across neuropsychiatric disorders and impact performance in this task (Alabi et al., 2019), we compared choice patterns at un-signaled contingency switches, noting no statistically significant alteration in KO mice (Figure 1—figure supplement 1I). We further probed cognitive flexibility with extra-dimensional set-shifting and spatial reversal tasks, again observing no performance differences between genotype (Figure 1—figure supplement 1J and K). In sum, choice abnormalities in Nrxn1α KO mice arise from decreased sensitivity to beneficial outcomes as opposed to altered feedback integration or impaired cognitive flexibility.

Neurexin1α mutants exhibit abnormalities in outcome-related task engagement

The temporal relationship between action and reinforcement modulates the degree to which rewards shape behavior. To assess whether observed differences in outcome sensitivity resulted from divergent temporal patterns of performance, we compared task latencies. We observed no significant discrepancies in latency to initiate between Nrxn1α wild-type and KO mice across varied reward environments (Figure 2A), suggesting that observed outcome-associated choice is not attributable to global task disengagement. Recent evidence suggests local choice value can also modulate the vigor with which selected actions are performed (Bari et al., 2019; Hamid et al., 2016). If inefficient choice patterns of Nrxn1α KOs result from disrupted value encoding, we expect that the effects of recent outcomes on action vigor would be similarly blunted. To explore this, we compared outcome-dependent initiation latencies after large reward versus small reward outcomes (Figure 2B). Interestingly, the relative latency to initiate trials in wild-type animals was significantly modulated by the relative reward ratio (Figure 2C, gray), with animals initiating trials more quickly after large reward outcomes than small reward outcomes. In contrast, Nrxn1α knockout mice were entirely unable to modulate initiation latency in response to the magnitude of previous reward (Figure 2C, blue). The strong inverse correlation between relative reward-stay and initiation latency was lost in Nrxn1α KO mice (Figure 2D and E). Thus, while there is no difference in average task latencies between wild types and KOs, Nrxn1α mutations disrupt outcome-modulated task engagement. We also observed a fixed elongation of choice latency in Nrxn1α mutants across reward environments (Figure 2—figure supplement 1A).

Figure 2 with 1 supplement see all
Neurexin1α mutants display altered outcome-dependent task engagement.

(A) A proxy of task engagement was measured as the average latency from trial onset (center-light ON) to initiation. Nrxn1α KOs (blue, n = 10) do not exhibit global deficits in task engagement in comparison to WT animals (gray, n = 11) (three-way RM ANOVA). (B) Relative latency to initiate is a standardized comparison of initiation latencies following large rewarded outcomes and small rewarded outcomes within individual animals. (C) Nrxn1α WT mice modulate their trial-by-trial engagement in response to different rewarded outcomes, initiating trials more quickly after large reward outcomes than small reward outcomes. Nrxn1α KOs do not exhibit this outcome-sensitive modulation of task engagement (three-way RM ANOVA). (D and E, top) There is a significant relationship between the ability of WT mice to select actions in response to reward discrepancy (RRS) and their ability to upregulate task engagement (relative initiation latency) which is lost in KOs (D and E, bottom). All data represented as mean ± SEM.

Value processing abnormalities in the Neurexin1α mouse extend to cost-based decision making

To see whether choice behavior based on costs was similarly affected in Nrxn1α mutants, we associated two choice alternatives with distinct motor requirements (fixed ratio 3 [FR3] vs. FR1; Figure 3A). Reward contingencies in this paradigm were not alternated and after 75 trials of feedback, mice achieved a steady-state response pattern. Interestingly, Nrxn1α KO mice do not select low-effort alternatives as frequently as wild-type littermates, both during sampling and steady-state periods (Figure 3B). While we noted the KOs slowed more over the session (Figure 3B), no significant difference in steady-state task engagement was seen (Figure 3C). We continued to observe an effect of genotype on choice latency (Figure 3D) as in prior tasks.

Neurexin1α mutants display a deficit in the selection of actions based on costs.

(A) Effort paradigm schematic. Mice distribute choices in a session with fixed contingency lasting 150 trials. Animals were given choices with equal reward outcomes, but different effort requirements (FR3 vs. FR1). (B) Nrxn1α KOs (blue, n = 10) choose less costly alternatives at a lower rate than their WT littermates (gray, n = 11) (two-way RM ANOVA). The distribution of choice in both WT and KO mice is altered over the course of the block as mice acquire information about the reward contingency, with a stable difference observed over the final 75 trials (two-sample t-test *p=0.023). (C) Nrxn1α KOs exhibited a clear interaction between trial and latency to initiate, slowing as they performed more high effort trials (two-way RM ANOVA). Nevertheless, there was no statistically significant difference in engagement at steady state (two-sample t-test p=0.14). (D) The longer choice latencies previously described in Nrxn1α KOs was observed in steady-state responding (two-way RM ANOVA; two-sample t-test *p=0.017). All data represented as mean ± SEM.

Reinforcement modeling reveals genotype-specific deficits in updating of outcome value

To uncover core decision-making processes underlying outcome-insensitive choice behavior in Nrxn1α mutants, we modeled action selection as a probabilistic choice between two alternatives with continually updating values (Figure 4A). We employed a modified Q-learning model with softmax decision function, including five parameters: (1) learning rate (α), which determines the extent to which new information about state-action pairing alters subsequent behavior; (2) reward compression parameter (γ), capturing the subjective benefit of a given reward volume; (3) inverse temperature parameter (β) ), linking the values of each option to choice output; (4) perseveration parameter (κ), capturing the effect of previous choices on subsequent choice, and (5) constant terms to capture spatial biases in choice behavior (see Materials and methods) (Doya, 2007; Niv, 2009; Vo et al., 2014).

Figure 4 with 1 supplement see all
A deficit in value updating underlies abnormal allocation of choices in Neurexin1α mutants.

(A) Q-learning reinforcement model. Mouse choice was modeled as a probabilistic choice between two options of different value (QL,QR) using a softmax decision function. Data from each reinforcement rate were grouped before model fitting. (B) Example of model prediction versus actual animal choice. Choice probability calculated in moving window of 13 trials. Long and short markers indicate large and small reward outcomes. (C and D) As compared to littermate controls (gray, n = 11), Nrxn1α mutants (blue, n = 10) exhibit a deficit in the learning rate, α, which describes the weight given to new reward information and γ, a utility function that relates how sensitively mice integrate rewards of different magnitudes (two-way RM ANOVA). (E) Nrxn1α KOs exhibit an enrichment of low ΔQ-value trials. (F and G) Nrxn1α mutants do not exhibit significant differences in explore–exploit behavior (F, captured by β) or in their persistence toward previously selected actions (G, captured by κ). (K) There is no significant difference in the decision function of Nrxn1α wild-type and mutant animals. All data represented as mean ± SEM. Bias figures can be found in Figure 4—figure supplement 1.

We have previously demonstrated stable trait-like reward processing characteristics in this task (Alabi et al., 2019). In light of this, we grouped the choice data of individual animals across reward ratios to extract stable behavioral parameters. We fit our model using function minimization routines and found that it provided accurate predictions of individual animal choice patterns (Figure 4B). Fitting choice data for wild-type and KO mice, we demonstrated that Nrxn1α KO mice have significantly lower α and γ parameters (Figure 4C and D), suggesting a global deficit in the updating and representation of choice values guiding decisions (Figure 4E). In contrast, we did not observe genotypic differences for the β, κ, or bias parameters (Figure 4F and G and Figure 4—figure supplement 1), suggesting no systemic differences in how the genotypes transform value representations into actions (Figure 4H).

Ablation of Neurexin1α in telencephalic projection neurons recapitulates value-based abnormalities

We next sought to identify molecularly causal circuits relevant for the deficits in value updating exhibited by Nrxn1α KO mice. Multiple telencephalic excitatory regions, which exhibit high expression of Nrxn1α mRNA, have been implicated in the regulation of action–outcome association and encoding of subjective choice value (Bari et al., 2019; Euston et al., 2012; Noonan et al., 2011; Padoa-Schioppa and Conen, 2017; Rushworth et al., 2011; Rushworth et al., 2012). To test whether Nrxn1α loss-of-function in these circuits could drive reward processing deficits, we crossed a Neurexin1α conditional allele (Nrxn1αfl), where exon 9 is surrounded by loxP sites, to the Nex-Cre transgenic line, where Cre-recombinase is driven from the Neurod6 locus in postmitotic progenitors of cortical, hippocampal, and amygdalar projection neurons (Goebbels et al., 2006; Figure 5A and B). mRNA from cortical dissection of Nrxn1αfl/fl; NexCre/+ revealed a 3.5× decrease in Nrxn1α transcripts spanning exon 9 as compared to Nrxn1αfl/fl; Nex+/+ (Figure 5C, left), and a modest degree of nonsense-mediated decay with a downstream probe (Figure 5C, right). Given the early expression of Cre from the NexCre/+ line, it is likely that the Nrxn1αfl allele is recombined prior to its endogenous expression (Lukacsovich et al., 2019). We choose this early deletion so as to best model the pathophysiological processes secondary to Nrxn1α mutations and make direct comparison to the phenotypes observed in the constitutive Nrxn1α KO mice.

Figure 5 with 1 supplement see all
Restricted telencephalic excitatory neuron deletion of Neurexin1α recapitulates choice abnormalities of constitutive KO.

(A) Nrxn1α was conditionally inactivated in telencephalic excitatory neurons by crossing a Nrxn1α-conditional knockout allele onto NexCre line. Controls for both the Nex (light gray) and Neurexin1α-conditional (dark gray) allele were analyzed. (B) Coronal section of brain from NEXCre/+;Ai14 (LSL-tdTOM) reporter cross showing restriction of tdTOM fluorescence to cortex, hippocampus, and a subdomain of the amygdala. (C) RT-qPCR of RNA from adult mouse cortex (n = 3 for Nrxn1αfl/fl; Nex+/+[dark gray] and Nrxn1αfl/fl; NexCre/+[purple]). Cre-mediated recombination results in reduced expression of Nrxn1α mRNA detected by exon 9 probe (two-sample t-test: p<0.0001) and moderate nonsense-mediated decay (two-sample t-test: p<0.01). (D) Nrxn1αfl/fl; NexCre/+ mutant animals (purple; n = 13) exhibit a reduction in relative reward-stay as compared with Nrxn1αfl/fl; Nex+/+(dark gray; n = 14) and Nrxn1α+/+; NexCre/+ (light gray; n = 11) controls. No difference in choice allocation was observed between control animals (genotype: p=0.88, relative reward: ***p<0.0001, probability: p=0.26, three-way interaction: p=0.25; three-way RM ANOVA). (E and F) Similar to Nrxn1α constitutive knockouts, Nrxn1αfl/fl; NexCre/+ mutant mice have a deficit in utilizing new reward information to update and represent choice values. The mutants exhibit a deficit in the learning rate (α) and in the reward volume sensitivity parameter (γ) (both analyzed by two-way RM ANOVA). (G) This leads to an enrichment of low ΔQ-value trials in mutant mice. (H–J) Nrxn1αfl/fl; NexCre/+ mutants do not differ from littermate controls for the relationship between choice value and decision behavior (H) and biases toward previous choice behavior (I). As a result, there is no significant difference in the decision function of control and mutant animals. (K–M) Nrxn1αfl/fl; NexCre/+ mutants exhibit a deficit in the allocation of choices guided by relative choice costs (K, two-way RM ANOVA, left; one-way ANOVA w/Tukey’s multiple comparison, right, *p<0.05). Mutants exhibit no difference in task engagement (L, one-way ANOVA w/Tukey’s multiple comparison, p>0.05) but recapitulate deficit in choice latencies (M, one-way ANOVA w/Tukey’s multiple comparison, **p<0.01). All data represented as mean ± SEM.

In order to test the effects of Nrxn1α loss-of-function in telencephalic projection neurons, we repeated the value-based tasks in Nrxn1αfl/fl; NexCre/+ mice. To account for potential hypomorphic effects of the Nrxn1α conditional allele as well as effects of constitutive Cre expression in the NexCre line, we utilized two controls: Nrxn1α+/+; NexCre/+ and Nrxn1αfl/fl; Nex+/+. We observed a significant effect of NexCre deletion of Nrxn1α on relative reward stay as compared to both control groups (Figure 5D). Similar to global Nrxn1α deletion, Nrxn1αfl/fl; NexCre/+ mutant animals were less able to bias their choice patterns toward more beneficial outcomes. We noted no consistent difference in behavioral flexibility in these mice (Figure 5—figure supplement 1A). Neither the Nrxn1αfl/fl; Nex+/+ conditional control nor the Nrxn1αfl/fl; NexCre/+ mutant animals displayed the reward-related modulation of initiation latencies observed in the Nrxn1α wild-type animals (Figure 5—figure supplement 1B), precluding conclusions regarding local modulation of action vigor. Similar to constitutive Nrxn1α KOs, we noted an increased choice latency across varied reward environments (Figure 5—figure supplement 1C). To test whether deficits in working memory could contribute to our choice phenotype, we assessed spontaneous alternation behavior of Nrxn1αfl/fl; NexCre/+ and Nrxn1αfl/fl; Nex+/+ conditional control littermates, observing no genotypic differences (Figure 5—figure supplement 1G).

To assess whether forebrain-specific Nrxn1α KOs generated similar reward processing abnormalities as Nrxn1α constitutive KOs, we again employed reinforcement modeling of choice data. As in whole-brain Nrxn1α KOs, we observed a significant effect of genotype on learning rate and reward discrimination parameters (Figure 5E and F), generating a leftward shift in the distribution of action value contrasts in Nrxn1αfl/fl; NexCre/+ mice (Figure 5G). In keeping with prior data, we observed no genotypic differences in value-related explore/exploit behavior, choice persistence, or average bias (Figure 5H–J and Figure 5—figure supplement 1D). In our effort-based cost paradigm, the Nrxn1αfl/fl; NexCre/+ conditional mutants exhibited reduced selection of the lower-cost alternative than both groups of control animals (Figure 5K). Average task engagement was not abnormal in these animals (Figure 5L and Figure 5—figure supplement 1E), but we again noted a persistent increase in choice latency (Figure 5M and Figure 5—figure supplement 1F). Together, these data suggest that embryonic deletion of Nrxn1α in telencephalic excitatory neurons is sufficient to produce similar perturbations of reward processing and choice as those observed in whole-brain Nrxn1α KO mice.

Deletion of Neurexin1α in thalamic nuclei does not recapitulate choice deficits

Neurexin1α is highly expressed in multiple subcortical regions involved in the selection and performance of goal-directed actions (Bradfield et al., 2013; Díaz-Hernández et al., 2018; Fuccillo et al., 2015; Ullrich et al., 1995). In order to assess the specificity of telencephalic excitatory Nrxn1α conditional KO (cKO) in driving reward processing abnormalities, we conditionally deleted Nrxn1α in developing thalamic nuclei via an Olig3-Cre driver line (Figure 6A–C). In contrast to telencephalic excitatory cKO, thalamic cKO could not recapitulate the deficits in value processing observed in whole-brain Nrxn1α mutants (Figure 6D and Figure 6—figure supplement 1A–C). There was no significant genotypic difference in the ability to modulate choice distributions in response to reward (Figure 6D), nor in any parameters of the fitted reinforcement model (Figure 6E–J and Figure 6—figure supplement 1D). Additionally, we noted no significant genotypic differences in choice allocation away from effortful alternatives (Figure 6K–M and Figure 6—figure supplement 1E). The only aspect of the constitutive KO phenotype partially recapitulated by the thalamic cKOs was increased choice latency in the fixed contingency paradigm (Figure 6—figure supplement 1F, but see Figure 6M).

Figure 6 with 1 supplement see all
Specific deletion of Neurexin1α in thalamic nuclei does not reproduce choice abnormalities observed in constitutive KO.

(A) Neurexin1α was conditionally inactivated in thalamic progenitor cells by crossing the Neurexin1α-conditional knockout line onto the Olig3-Cre line. (B) Coronal section of Olig3Cre; Ai14 reporter cross showing expression of tdTOM broadly throughout thalamic nuclei. (C) RT-qPCR of RNA from adult mouse thalamus (n = 2 for Nrxn1αfl/fl;Olig3+/+ (gray); n = 3 for Nrxn1αfl/fl;Olig3Cre/+(orange)). Cre-mediated recombination results in reduced expression of Nrxn1α mRNA detected by exon 9 probe (two-sample t-test: p<0.0001) and moderate nonsense-mediated decay (two-sample t-test: p<0.001) (D) Nrxn1αfl/fl;Olig3Cre/+ mutant animals (orange; n = 10) do not exhibit changes in relative reward-stay in comparison with Nrxn1αfl/fl;Olig3+/+(gray; n = 8) control animals. (E–G) Nrxn1αfl/fl;Olig3Cre/+ mutant mice do not have a deficit in updating or representing choice values (two-way RM ANOVA). (H–J) Nrxn1αfl/fl;Olig3Cre/+ mutants exhibit a normal relationship between choice values and decision behavior. (K–M) Nrxn1αfl/fl;Olig3Cre/+ mutants do not exhibit a deficit in the allocation of choices guided by relative choice costs (K, two-way RM ANOVA, left; two-sample t-test, right, p>0.05). Mutants exhibit no difference in task engagement (L, p>0.05) or in choice latencies (M, p>0.05). All data represented as mean ± SEM.

Characterizing value-modulated neural signals within dorsal striatum

Our data suggest that both global and telencephalic excitatory neuron-specific Nrxn1α mutants exhibit inefficient choice patterns secondary to deficits in value encoding/updating. Given the function of Nrxn1α in supporting excitatory synaptic transmission in hippocampal circuits (Etherton et al., 2009), we explored how its disruption might impact neural activity within key reinforcement learning circuits. We focused on direct pathway spiny projection neurons (dSPNs) of the dorsal striatum, as this population: (1) is a common downstream target of forebrain excitatory populations that both encode value and express Nrxn1α in their presynaptic terminals (Bari et al., 2019; Bradfield et al., 2013; Parker et al., 2019); (2) encodes reward values (Donahue et al., 2019; Samejima et al., 2005; Shin et al., 2018); and (3) can bias choice in value-based operant tasks (Kravitz et al., 2012; Tai et al., 2012). To select for striatal dSPNs, we expressed GCamp6f in neurons projecting to the substantia nigra reticulata (SNr), via combined injection of retroAAV2.EF1α−3xFLAG-Cre in the SNr and AAV5.hSyn-DIO-GCamp6f in the dorsal striatum of control NEXCre mice (Figure 7A and B). Putative direct pathway SPNs (p-dSPNs) exhibited reproducible Ca2+ activity patterns in relation to three task epochs – trial start (center port light on), self-initiation (center port entrance), and choice/reward delivery (side port entry) (Figure 7C), despite exhibiting smaller average signals than during task disengagement (Figure 7—figure supplement 1A). The lack of similar waveforms on the isosbestic 405 nm channel confirms the specificity of these epoch-aligned Ca2+ signals (Figure 7—figure supplement 1B).

Figure 7 with 1 supplement see all
Quantifying value correlates in putative direct pathway SPNs of the dorsomedial striatum.

(A) Schematic of experimental scheme. Control (Nrxn1α+/+; NexCre/+, n = 7) mice were injected with a retro-AAV2-EF1α−3xFLAG-Cre virus in the substantia nigra, pars reticulata (SNr). Ipsilateral injection of Cre-dependent GCamp6f allowed for enrichment of putative direct pathway SPNs (p-dSPNs). (B, top) Sagittal section of NexCre brain showing GCamp6f expression in dorsal striatal SPNs and placement of 400 µm optic fiber (white arrow). (B, bottom) Magnified view of striatum showing colocalization of nuclear FLAG-Cre and cytoplasmic GCamp6f. (B, bottom left) Location of fiber placements in NexCre/+. (C, top) Trial schematic and relationship of specific task epochs with p-dSPN Ca2+ signal (bottom). (D) Peristimulus time histogram (PSTH) of ΔF/F for NexCre/+ aligned to initiation event (all trials). The initiation of the action sequence (green bar) is associated with a rise in p-dSPNs activity. (E) Representative heat map of individual animal trials segregated by reward outcome on (t−1) trial (sorted by the latency to initiate). Trials following a large reward have greater signal suppression than those following small reward. (F) PSTH of ΔF/F for NexCre/+ aligned to initiation event (segregated by outcome on (t−1)). Preinitiation of p-dSPN dynamics exhibits two components – a slow ramping phase (yellow, time-10→-1) followed by a fast spike phase (green, time-1→init), both of which are modulated by (t−1) reward outcome. (G) The slow ramping phase is quantified by the integral of GCamp signal −10 s to −1 s before initiation. (H) There is a significant effect of (t−1) reward volume on the preinitiation integral during slow ramping with large rewards showing greater silencing of p-dSPN activity (paired t-test, ***p=0.0002). (I) Preinitiation integral inversely correlates with the comparative action value of the upcoming trial, which is calculated using probability estimates from fitted reinforcement learning models and reflects the disparity in choice value on a trial to trial basis. (J) The dynamics of the fast peak phase are represented by the average slope of GCamp signal from −1 s till initiation. (K) There is a significant effect of (t−1) reward volume on preinitiation slope during the fast peak phase (paired t-test,***p=0.0006) with large rewards showing steeper subsequent preinitiation slopes. (L) Preinitiation slope positively correlates with the comparative action value of the upcoming trial.

Recent population Ca2+ imaging of striatal SPN populations has revealed a prolonged ramping activity prior to action sequence initiation (London et al., 2018). Given our data (Figure 2) and other work documenting the modulation of initiation latency by prior outcome (Bari et al., 2019), in addition to the technical challenges of reliably separating the choice and outcome components of the Ca2+ waveform (Figure 7—figure supplement 1C–F), we investigated the preinitiation window as a key epoch for value-modulated signals in striatal direct pathway neurons. An average of all trials aligned by initiation demonstrated slow and fast phases of the p-dSPN Ca2+ waveform (Figure 7D). To understand how reward correlates with wild-type p-dSPN activity, we segregated trials by previous (t−1) outcome. We found that most pre-initiation epochs following a ‘small reward’ trial had elevated activity compared to the population Ca2+ average, while trials following ‘large reward’ had suppressed activity relative to the population average (Figure 7E), a trend similarly present in the population data (Figure 7F). To further quantify signal dynamics, we examined the slow ramping phase, occurring ~10 s before an initiation, and the fast peaking phase, occurring 1 s before initiation. We found that both signal components were differentially modulated by reward outcome: (1) for slow ramping, (t−1) large reward outcomes result in negative ramping or silencing of p-dSPN activity in comparison with small rewards (Figure 7G and H); (2) for fast peaking, larger rewards result in steeper peak activity as compared to smaller rewards (Figure 7J and K). Furthermore, we noted significant correlations between both measures and trial-by-trial comparative action values (Figure 7I and L; see Materials and methods), suggesting these p-dSPN signals may reflect value information employed for future action selection.

Neurexin1α deletion in excitatory telencephalic projection neurons disrupts value-associated striatal neuron activity

To examine whether deletion of Nrxn1α from telencephalic projection neurons disrupted value-modulated neural signals within striatum, we performed population Ca2+ imaging of p-dSPNs in both Nrxn1α+/+; NexCre/+ (Nex-Control) and Nrxn1αfl/fl; NexCre/+ (Nex-Nrxn1αcKO) mice during our serial reversal task. While we did not uncover a difference for the slow ramp signal component between genotypes (Figure 8A–C), we found that the slope of the fast peak was consistently lower in Nex-Nrxn1αcKO (Figure 8D and E). Furthermore, this deficit was specifically associated with failure to increase peak activity in response to large reward volumes (Figure 8F and G). To assure that our strategy for labeling d-SPNs, wherein Cre becomes expressed in the recorded spiny neurons, did not alter recurrent inhibition, we compared a separate set of Nex-Nrxn1αcKO mice injected with either retroAAV2.EF1α−3xFLAG-Cre or retroAAV2.hSyn-GFP-ΔCre (an enzymatically inactive truncated version of Cre) in the SNr and noted no difference in the frequency or amplitude of miniature inhibitory postsynaptic currents (mIPSCs) according to virus (Figure 8—figure supplement 1A and B). To rule-out any potential effects on excitatory striatal afferents, we performed a similar experiment on Nrxn1αfl/fl mice, again noting no difference in the miniature excitatory postsynaptic currents (mEPSCs) in retrograde Cre versus ΔCre viruses (Figure 8—figure supplement 1C and D). Together, these data suggest that telencephalic excitatory neuron-specific Nrxn1α mutants do not exhibit global disruptions of striatal circuit dynamics, but a specific outcome-associated perturbation in fast peak activity prior to trial initiation.

Figure 8 with 1 supplement see all
Restricted telencephalic excitatory neuron deletion of Neurexin1α produces a deficit in fast peak activity in p-dSPNs of the DMS.

(A and B) PSTH of ΔF/F for Nex-control (Nrxn1α+/+; NexCre/+, n = 7, gray) and Nex-Nrxn1αcKO (Nrxn1αfl/fl; NexCre/+, n = 6, purple) mice, respectively, aligned to initiation event (segregated by outcome on t−1). Shaded region corresponds to the difference in the preinitiation integral following large and small reward outcomes. (C) There is no statistically significant difference between Nex-control and Nex-Nrxn1αcKO in the Δpre-initiation integral of large versus small rewards (two-sample t-test, n.s., p=0.084). (D and E) PSTH of ΔF/F for control and mutant animals, respectively, in the fast peak phase of preinitiation activity. (F) Nex-Nrxn1αcKO exhibit smaller disparity in fast peak signals after unique reward outcomes, as evidenced by significant effect of genotype on Δpre-initiation slope of the fast peak (two-sample t-test, *p=0.025). (G) This difference in Δpre-initiation slope arises from a blunted GCamp response in mutants to large reward outcomes (two-way RM ANOVA). (H) Modeling Ca2+ signal dynamics as function of reward variables (blue), prior/future choice (gold), and lagging regressors (light blue) to capture prior circuit states. Value modulation of fast peak activity is blunted in Nex-Nrxn1αcKO mice (highlighted red box), while other components of the signal remain intact. Slow ramping is largely intact in mutant animals. All data represented as mean ± SEM.

To better understand whether mutation-associated changes in striatal neural signals are related to specific components of value-based decision making, we developed a linear-mixed effects model to explain variability in the preinitiation phases of p-dSPN signals. Our model included variables for reward processing (prior trial reward outcome and reward prediction error, disparity in action value between choices in the upcoming trial), choice behavior (choice, explore–exploit, and stay–shift strategies), task engagement (initiation latencies), and lagging regressors to reflect ‘carry-over’ effects from previous trials (Figure 8H, see Materials and methods). We found that blunting of fast peak dynamics in Nex-Nrxn1αcKO mutants was specific to aspects of reward processing – that is, while peak slopes had significant correlation to reward history, reward prediction error, and comparative choice values in wild-type mice, these outcome-sensitive signal components were absent in mutant striatal population dynamics (Figure 8H). In contrast, value-modulated signal components are preserved in the mutants during slow ramping (Figure 8H), supporting a circumscribed alteration in striatal value coding. Together, these data demonstrate disrupted reward responsive activity in direct pathway SPNs upon ablation of Nrxn1α in a subset of excitatory forebrain neurons. These changes are broadly consistent with our behavioral analysis showing Nrxn1α knockout in frontal projection neurons produced lower learning rate and sensitivity to outcome magnitudes (Figure 5E and F), generating smaller Q value discrepancies (Figure 5G).

Discussion

Understanding genetic contributions to brain disease requires bridging the sizeable chasm between molecular dysfunction and behavioral change. While behaviorally circumscribed neural circuits provide a logical intermediary substrate, it has been challenging to identify disease-relevant neural populations owing to: (1) difficulty in finding assays that provide stable readouts of relevant behavioral constructs; (2) incomplete understanding of specific computational algorithms and neural circuit implementations for behavioral constructs; (3) challenges localizing relevant neural circuits wherein gene perturbations drive behavioral dysfunction; and (4) limitations in correlating mutation-associated patterns of neural activity with abnormal execution of behavior.

Here we addressed these obstacles while investigating value-processing deficits in mice harboring mutations in Nrxn1α, a synaptic adhesion molecule associated with numerous neuropsychiatric disorders (Dachtler et al., 2015; Duong et al., 2012; Huang et al., 2017; Kirov et al., 2009; Rujescu et al., 2009; Sanders et al., 2015; Südhof, 2008). We found that constitutive Nrxn1α KO mice exhibited reduced bias toward more beneficial outcomes (modeled by greater reward volumes) and away from more costly actions (modeled by higher response schedules). Reinforcement modeling of choice behavior suggested altered mutant decision making resulted from deficits in the updating and representation of choice value as opposed to how these values are transformed into action. Using brain region-specific gene manipulation, we demonstrated that deletion of Nrxn1α from telencephalic projection neurons, but not thalamic neurons, was able to recapitulate most aspects of the reward processing deficits observed in constitutive Nrxn1α KOs. Finally, we investigated how circuit-specific Nrxn1α mutants altered value-modulated neural signals within direct pathway neurons of the dorsal striatum. We found that while fast peak Ca2+ activity immediately preceding trial initiation strongly reflected aspects of prior and current action values in wild-type mice, value-coding signals were disrupted in telencephalic-specific Nrxn1α mutants.

Deficits in value-based action selection in Neurexin1α mutants

Reframing the study of disease-associated behaviors into endophenotypes is a powerful approach to revealing underlying genetic causality. Nevertheless, the study of disease-relevant cognitive endophenotypes in mice has proven challenging. Here we employed a feedback-based, two-alternative forced choice task that forces value comparisons between choices of differing reward magnitude and required effort. We believe this task has many advantages for investigating cognitive dysfunction associated with neuropsychiatric disease risk genes such as Nrxn1α. First, we have previously shown that it produces stable within-mouse measures of benefit and cost sensitivity (Alabi et al., 2019), ideal for revealing between-genotype differences. Second, it probes how outcome value is used to direct future action selection – a core neural process perturbed across many of the brain disorders in which Nrxn1α mutations have been implicated (Dichter et al., 2012; Gillan and Robbins, 2014; Maia and Frank, 2011).

We find that global deletion of Nrxn1α resulted in a persistent deficit in outcome-associated choice allocation, driven strongly by reductions in win–stay behavior (Figure 1C–E). Interestingly, similar reductions in win–stay behavior during feedback-based tasks have been demonstrated to drive choice inefficiency in both schizophrenia (Saperia et al., 2019) and autism (Solomon et al., 2015), disorders for which Nrxn1α has been implicated. We observed that this value-based dysregulation manifests not only for the selection of higher-benefit actions, but also in the selection of less costly choices (Figure 3), as well as in the outcome-dependent modulation of task engagement as read out by initiation latency (Figure 2). Together, these data converge to suggest Nrxn1α mutations disrupt the function of brain circuits that internally represent value or circuits that transform these encoded values into actions.

Deficits in the updating and representation of value are core computational deficits in Neurexin1α mutants

In order to reveal which aspects of the decision process were altered in Nrxn1α mutants, we took advantage of Q-learning models to quantitatively describe relevant drivers of choice in feedback-based reinforcement paradigms (Daw, 2011; Sutton and Barto, 1998). Our data suggest that choice abnormalities in Nrxn1α KO mice reflect deficits in the updating or encoding of choice values, encapsulated by reductions in the learning rate (α) and outcome differentiation (γ) parameters, as opposed to differences in how mice translate value into action (β) or persist on actions independent of outcome (κ) (Figure 4). These data are reminiscent of work from schizophrenic subjects in a probabilistic reinforcement learning paradigm, where similar modeling suggested a reduction in the learning rate in patients versus neurotypical controls (Hernaus et al., 2018). Of particular interest, these investigators interpreted alterations in learning rate not to reflect perturbations in the reward prediction error (RPE) signal itself but to changes in how those signals were integrated to update value for future actions (Hernaus et al., 2019; Hernaus et al., 2018). While we cannot directly map parameters of the reinforcement model to neural circuits, this interpretation suggests that relevant circuit loci might be those tasked with integrating dopaminergic RPE signals, including connections between cortical regions and the striatum.

Deletion of Neurexin1α from telencephalic excitatory neurons recapitulates choice abnormalities of the constitutive knockout

The above hypothesis, together with robust expression of presynaptically expressed Nrxn1α throughout cortex and its known role in mediating excitatory synaptic function in hippocampal circuits, directed us toward probing its function in corticostriatal circuits. A large literature has implicated multiple excitatory forebrain populations in flexibly encoding the expected value of anticipated reward (Kennerley and Walton, 2011; Rolls, 2000; Tremblay and Schultz, 1999; Wallis and Kennerley, 2010; Wallis and Miller, 2003), reward-dependent modulation of working memory (Wallis and Kennerley, 2010), and forming associations between motivated behaviors and their outcomes (Hayden and Platt, 2010). Consistent with this, deletion of Nrxn1α from embryonic telencephalic excitatory neuron progenitors recapitulated the value-based deficits observed in the constitutive KOs (Figure 5). While we do not claim this as the sole circuit-specific deletion capable of generating this phenotype, some degree of specificity was demonstrated by the absence of decision-making phenotypes in our thalamic Nrxn1α deletion (Figure 6).

Unfortunately, the broad recombinase expression of the Nex-Cre transgenic within telencephalic excitatory populations precludes us from assessing the importance of Nrxn1α in specific telencephalic populations such as medial prefrontal or sensorimotor cortices. It also cannot rule out a role for excitatory populations within the amygdala that have been linked to goal-directed instrumental actions (Corbit et al., 2013). Co-expression networks seeded by autism candidate genes have highlighted human mid-fetal deep layer cortical neurons from both prefrontal and primary motor/somatosensory cortices as potential sites of autism pathogenesis (Willsey et al., 2013). Furthermore, human patients with damage to the ventromedial prefrontal cortex exhibit similar deficits in value-based decision-making tasks as those seen in our Nex deletions (Camille et al., 2011; Fellows and Farah, 2007). Further assessment of the contribution of prefrontal Nrxn1α function to the observed phenotypes awaits Cre transgenic lines with both greater cortical regional specificity and embryonic expression. It is worth noting that Nex-Cre transgenic mice also label a small subset (~10%) of VTA neurons that project to the medial shell of the nucleus accumbens (Bimpisidis et al., 2019; Kramer et al., 2018). While we cannot formally rule out the contribution of these neurons to our behavioral results, they are unlikely to account for our Ca2+ imaging results, as their projections are distant from our imaging site.

Circuit-specific ablation of Neurexin1α disrupts value-modulated neural signals within striatum

Based on our behavioral data and computational modeling from multiple Nrxn1α mutants, expression patterns of Nrxn1α transcripts (Fuccillo et al., 2015), and the known pre-synaptic function of this molecule in maintaining synaptic connectivity (Anderson et al., 2015; Aoto et al., 2013; Etherton et al., 2009; Missler et al., 2003), we hypothesized that the observed value-based abnormalities resulted from altered synaptic transmission at key sites for integration of RPEs into action value coding. Putative circuit loci include: (1) connections within value-encoding forebrain excitatory areas; (2) connections from cortex onto mesencephalic dopamine neurons that encode striatal-targeting RPE signals (Takahashi et al., 2011); and (3) connections from cortical areas into striatum. Reasoning the aforementioned possibilities would all impact neural signals of striatal SPNs, we recorded population Ca2+ activity of putative dSPNs via fiber photometry (Figure 7A–C). In support of this idea, we observed value-modulated signals leading up to trial initiation (Figures 7D and F and 8A,D, and H), consistent with population Ca2+ imaging signals observed in both SPN subtypes as mice approach palatable food (London et al., 2018). While our imaging does not provide the clarity of cellular-level approaches (Donahue et al., 2019; Kwak and Jung, 2019), it clearly resolved two phases of activity – a slow ramp occurring ~10 s before trial initiation and a fast peak in the 1 s leading up to initiation – that correlated with prior reward outcome and RPE (Figures 7 and 8). Interestingly, the Nex-Nrxn1α mutants displayed a clear disruption of these reward variable correlations with p-dSPN activity, specifically for the fast peak immediately preceding trial initiation (Figure 8H). We suggest a hypothesis wherein RPE signals are not appropriately integrated in Nex-Nrxn1α mutants, depriving striatal circuits of essential reward relevant information for subsequent action selection (Hernaus et al., 2019; Hernaus et al., 2018). Recent evidence from ex vivo brain slices suggests complex alterations to excitatory synaptic transmission for both anterior cortical and thalamic projections to striatum (Davatolhagh and Fuccillo, 2020). Nevertheless, in vivo neural recordings of both cortico-striatal and value-encoding cortical circuits during this task will be needed to understand how Nrxn1α mutations contribute to altered striatal representations of value.

Extensive associations have been found between mutations in Nrxn1α and a range of neuropsychiatric disorders (Dachtler et al., 2015; Duong et al., 2012; Huang et al., 2017; Kirov et al., 2009; Rujescu et al., 2009; Sanders et al., 2015; Südhof, 2008). Here we show that Nrxn1α plays a key functional role in specific forebrain excitatory projection circuits governing cognitive control of value-based action selection. It is interesting to speculate that the widespread nature of basic reinforcement learning abnormalities seen across neuropsychiatric diseases could be explained by similar network dysfunctions as seen here for Nrxn1α mutants. Further work will be necessary to test the generalizability of these observations for other neurodevelopmental psychiatric disorders and further refine the telencephalic excitatory populations of relevance.

Materials and methods

Contact for reagent and resource sharing

Request a detailed protocol

Code used for data analysis is available on the public Fuccillo lab github site (https://github.com/oalabi76/Nrxn_BehaviorAndAnalysis; Alabi, 2020; copy archived at swh:1:rev:b8233aab4e607f82c868caf2dfe4007790088e8e). Data for this manuscript is posted to Dryad (Alabi et al., 2019, Neurexin Photometry, Dryad, Dataset, https://doi.org/10.5061/dryad.vhhmgqnrq). Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Marc Fuccillo (fuccillo@pennmedicine.upenn.edu).

Experimental model and subject details

Request a detailed protocol

Animal procedures were approved by the University of Pennsylvania Harbor Laboratory Animal Care and Use Committee and carried out in accordance with National Institutes of Health standards. Constitutive Neurexin1α (Nrxn1α) KO mice were obtained from the Südhof lab (Stanford University) (Geppert et al., 1998). Nrxn1α+/- males and females were bred to produce subject for this study. In sum, 11 Nrxn1α+/+ and 12 Nrxn1α-/- mice were used in this study. One Nrxn1α-/- mouse died in the early stages of training and its results were excluded. Nrxn1α conditional knockout mice were generated from sperm stock (Nrxn1 <tm1a(KOMP)Wtsi>) heterozygotes on the (C57Bl/6N background) obtained from the MRC Mary Lyon Center (Harwell, UK). The lacZ gene was removed via crosses to a germline-FLP recombinase, which was then bred off, followed by at least four generations breeding to homozygosity within our colony. NexCre mice (kind gift of Klaus-Armin Nave and Sandra Goebbels, Göttingen, Germany) were obtained and crossed onto Nrxn1αc/c mice (Goebbels et al., 2006). In this study 11 Nex+/- Nrxn1α-/-, 14 Nex-/- Nrxn1αc/c, and 13 Nex+/- Nrxn1αc/c mice were used. Olig3Cre mice were obtained (kind gift of Yasushi Nakagawa, University of Minnesota) and similarly crossed onto the Nrxn1αc/c colony (Vue et al., 2009). In this study 8 Olig+/+; Nrxn1αC/C and 10 OligCre/+; Nrxn1αC/C mice were used.

Whenever possible, animals were housed in cages with at least one littermate. One Neurexin1α wild-type and two Neurexin1α knockout animals were singly housed to avoid injury from fighting. Mice were food-restricted to maintain 85–90% of normal body weight and were given ad libitum access to water throughout the duration of the experiment. Mice were allotted 0.2–0.4 g of extra food on non-experimental days to account for the discrepancy in caloric intake from not receiving reward in a task. A 7 AM to 7 PM regular light–dark cycle was implemented for all mice used in this study. Cages were maintained in constant temperature and humidity conditions.

Behavioral apparatus and structure

Request a detailed protocol

Experiments were conducted utilizing Bpod, a system specialized for precise measurements of mouse behavior (Sanworks LLC, Stony Brook, NY). A modular behavioral chamber (dimensions 7.5 L × 5.5 W × 5.13 H inches, ID: 1030) with three ports capable of providing light cues and delivering liquid rewards was used to measure behavioral events. Each port was 3D printed from clear XT Copolyester and housed an infrared emitter and phototransistor to measure port entries and exits precisely. Behavior chambers were enclosed in larger sound-attenuating boxes. For each behavioral paradigm, illumination of the center port after a 1 s intertrial interval indicated the beginning of a trial. Animals initiated trials by registering an entry to the lit center port, triggering a choice-period. The choice period was marked by the extinction of the center light and illumination of the ports on either side of the center. Mice were given an x-sec (varied by protocol) temporal window to enter either the left or right port and register a choice. Failure to register a choice in this period resulted in an omission, which was followed by a 3 s timeout and required the animal to reinitiate the task.

Successful registration of a choice resulted in the extinction of all port lights and the delivery of a variable volume of liquid supersac reward (3% glucose, 0.2% saccharin in filtered water) via a steel tube in the choice ports. Reward volumes and delivery probabilities were dependent on task conditions. The reward period lasted a minimum of 5 s. Following this mandatory minimum, the reward phase was extended if a mouse was noted to be occupying one of the three ports. The trial ended only after successful confirmation of port exit from all three ports. Reward volumes were regulated via individually calibrated solenoid valves, with specific time/volume curves to deliver precise reinforcement.

All port entries, exits, and other task events were recorded by the Bpod State Machine R1 (ID: 1027) and saved in MATLAB. Behavioral protocols and primary analysis were developed in MATLAB.

Operant behavior

Request a detailed protocol

Acquisition of goal-directed contingency

Request a detailed protocol

Mice were habituated to behavior chambers and ports over a 3-day period. Each day, animals were given a 10 min adjustment period followed by a program delivering 10 µL of reward every 30 s for 40 min. The first 40 trials were grouped into two blocks, with reward delivered either from the left or the right port for 20 contiguous trials. Following this period, reward was alternated between left and right port for the remaining 20 trials. Port lights were illuminated for a 10 s period to indicate reward delivery, followed by a 20 s ITI.

Following this introductory period, mice were introduced to a goal-directed task that required them to acquire a light-chasing reward contingency. Trials were initiated as described previously. During the choice phase, one of the two lateralized ports was illuminated at random. Mice were given 10 s to register a choice, or an omission was charged. If entries into the unlit lateral port or the center port were registered a 3 s timeout occurred and the animals had to reinitiate the trial until they selected the correct port. Successful selection of the correct port resulted in 10 µL of reward (Prew = 1.0). Sessions lasted 1 hr with no trial number limits. After 10 sessions, mice that had completed two consecutive days of >125 trials or 1 day >200 trials progressed to the serial reversal task. If mice missed this deadline, they were again assessed after their twelfth session. No mice failed to meet these criteria by the twelfth session.

Serial reversal value task

Request a detailed protocol

After successfully acquiring the action–outcome contingency described above, mice progressed to a forced-choice two-alternative serial reversal paradigm with variable reward outcomes. Trial initiation occurred as described above, via entry into the central port. To ensure accurate initiation latencies, the state of the center port was assessed after the ITI. The beginning of a trial was delayed if a mouse was found occupying this port. Initiation of a trial led to a 5 s choice period in which both left and right lateral ports were illuminated as choice alternatives. Following selection, a variable volume of reward was delivered contingent upon current task conditions (Prew = 0.75 and 0.4 were used here). The reward phase lasted 5 s and trial termination did not occur till after mice successfully disengaged from all ports. One Nrxn1α-/- mutant animal was excluded from the reversal study due to mis-calibrated solenoid valves.

Similar to our previous study, a ‘moving window’ of proximal task events was used to monitor mouse choice patterns (Alabi et al., 2019). Changes of choice-outcome contingencies were initiated when 8 of the last 10 actions were allocated to the large reward volume side. Following detection of this event, the lateralization of reward volumes was switched. These contingency reversals were un-cued and served to mitigate outcome-insensitive behavior. Reward probabilities were the same for both choices and consistent over a given session. The relative reward contrast was consistent over a given session. Eight reward environments were tested (four relative reward ratios across two reward reinforcement rates). Animals performed the eight tests in a random sequence, performing the high reinforcement sessions before the low reinforcement sessions. For initial introduction to task structure, mice were run in the reversal paradigm (12 µL vs. 0 µL) for 5–8 days prior to initiating the sequence of behaviors described above. All sessions were limited to 1 hr with no cap on trial number. Reward, however, was limited to 2000 µL in a session.

To ensure that behavioral measures were not overly influenced by spatial bias developed in one session (which could last for many subsequent sessions, across reward environments), sessions with excessive or carryover bias were excluded from this study and triggered a re-training phase before the experiment was continued. Bias was calculated as:

Overall Bias= (Pokes(Bias)Pokes(Non-Bias))/Total\ Pokes

where Pokes (Bias) denotes the number of port entries to the side which received more pokes and Pokes (Non-Bias) represents the number of pokes to the side that received less. A bias exceeding 0.45 initiated an automatic re-training phase lasting at least one session. Sessions with biases >0.2 triggered a watch-period in mice. If another session produced a bias >0.2 to the same spatial choice alternative, that session was marked as having carry-over bias from a previous session and excluded – also triggering a retraining phase. Sessions were additionally excluded if animals met three conditions in a single session: (1) overall bias exceeding 0.45; (2) failure to complete a minimum of two contingency switches; and (3) failure to complete at least 100 selections of the nonbiased alternative. During re-training, animals performed one session of the 12 µL vs. 0 µL reversal task to eliminate spatial bias.

Static contingency effort task

Request a detailed protocol

A behavioral paradigm with a stable reward contingency over 150 trials was used to assess how costs shape behavior. Cost was modeled as increased operant responding (FR3) before delivery of a reward. Costs were applied to one alternative for 150 trials, following which a relative reward reversal was initiated (10 µL vs. 0 µL) to eliminate the spatial bias developed during the task. Entry into one port during the choice phase led to extinction of the contralateral light. The chosen port remained lit until the animal completed the repetitive motor requirement necessary to obtain reward. Immediately upon completion of this requirement, reward was delivered as described previously. Equal reward volumes (8 µL, Prew = 1) were implemented during the experimental phase of this task. Trial structure was the same as in the reversal paradigm described above. All sessions were limited to 1 hr. Each animal performed two experimental sessions to account for potential spatial biases. One with the high motor threshold on the right and the other with it on the left choice port.

Before animals were exposed to relative costs, they were acclimated to the new behavioral requirements by a three-session minimum training period in which they completed this task with an FR3 vs. FR3 to increase response rate.

Cognitive flexibility assays

Request a detailed protocol

To measure cognitive flexibility, we employed an attentional set shifting task where the correct port was first indicated by a lit visual cue and subsequently switched to a fixed egocentric spatial position. Trials were structured as previously described. In the first 25 trials, a light cue denoted the position of reward. Mice initiated trials in which one of the lateralized alternatives was illuminated, at random, during a 10 s choice window. Selection of the illuminated port resulted in a 10 µL reward, and selection of the unlit port resulted in a timeout. Following this baseline block, illumination of the choice ports continued to occur at random, but rewards were only delivered on one of the choice ports for the remainder of the session. Sessions were capped at 1 hr and 250 trials.

To further probe behavioral flexibility, we utilized an egocentric spatial reversal task. Individual trial structure was preserved. In the first block of 25 trials, one of the choice ports was assigned as the reward port. Following this introductory block, the opposite port was assigned as the reward port. On each trial, one of the two ports was illuminated at random. A 10 µL reward was given after selection of the appropriate port.

To account the potential biases and intersession fluctuations in performance, each animal was tested twice in each behavior – with alternating spatial cues in each session. Prew = 1 for both behaviors upon selection of correct alternative.

Spontaneous alternation behavior

Request a detailed protocol

Mice were acclimatized to the testing room for 1 hr prior to testing. Alternating behavior was measured in a Y-maze (custom built, based on San Diego Instruments Y-maze 2005) and recorded with an overhead camera (10fps). To begin the test, each mouse was placed in arm C facing arms A and B. The mouse was allowed to freely explore the Y-maze for 5–8 min. If the mouse performed 15 arm entries (defined as entry of all four limbs into an arm) by the end of 5 min, the session was ended immediately. If the mouse had not performed 15 arm entries after 5 min, an additional 3 min was given. Mice that did not perform 15 arm entries within 8 min were excluded from the data. The video was manually scored by an experimenter who was blinded to the animal's genotype and sex.

Analysis of behavioral performance

Request a detailed protocol

Data were analyzed using custom-written scripts developed in Matlab (R Development Core Team, 2017). We utilized basic function supplemented by the following toolboxes: Bioinformatics, Curve Fitting, Data Acquisition, Global Optimization, Parallel Computing (R Development Core Team, 2017). Analytical code is available on request.

Descriptive parameters

Request a detailed protocol

The session performance index was calculated as:

PerformanceIndex=eln(Pr(Large Reward)1Pr(Large Reward))

where Pr(Large Reward) refers to the percentage of total choice that animals made to the large reward alternative over the course of a session.

The relative reward-stay of an outcome, A, versus another outcome, B, was calculated as:

RelativeRewardStay=ln((Pr(A)1Pr(A))/(Pr(B)1Pr(B)))

where Pr(A) and Pr(B) refer to the probability that mice stay on the choice alternative producing outcome A and B, respectively, on the t−1 trial.

The adaptability index was calculated as:

AdaptabilityIndex=(i=1n((LipostSipost)+ (LipreSipre))/10)/n

where Lipre and Lipost refer to the number of large alternative selections in the 10 trials before and after the i-th contingency switch in an individual session and Sipre and Sipost refer to the number of small alternative selections in the same time window. n is the number of blocks completed in a session.

The relative initiation latency was calculated as:

RelativeLatencytoInitiate= (LatInitLargeLatInitSmall)/LatInitSmall

where LatInitLarge and LatInitSmall refer to the average latency to initiate trials following large reward and small reward outcomes, respectively, in an individual session.

Logistic regression

Request a detailed protocol

We employed a logistic regression to model current choice as a function of past actions and outcomes (n = 5 trials):

log(R(i)1R(i))=β0+p=1nβpLRLR(ip)+p=1nβpSRSR(ip)+p=1nβpNRNR(ip)+p=1nβpCC(ip)+error

where R(i) is the probability of choosing the right-sided alternative on the ith trial. LR(i−p), SR(i−p), and NR(i−p) refer to the outcomes of the pth trial before the ith trial. LR(i−p) is defined such that LR(i−p) = +1 if an animal received a large reward resulting from a right press on the pth previous trial, −1 if an animal received a large reward resulting from a left press on the pth previous trial, and 0 if the animal did not receive a large reward on that trial. SR(i−p) and NR(i−p) are defined similarly for trials that resulted in small reward and no reward outcomes, respectively. C(i−p) is an indicator variable representing the previous choice behavior of the mouse (C = 1 for right-sided choice and C = 0 for left-sided choice). These variables provide a complete accounting of the choice, reward history, and interaction of the two in our task. This method assumes equivalent reinforcement from outcomes regardless of the lateralization of choice. The model was fit to six random blocks of 85% of choice data. The coefficient produced by these blocks was averaged to produce individual coefficients for each animal. Regression coefficients were fit to individual mouse data using the glmfit function in Matlab with the binomial error distribution family. Coefficient values for individual mice were averaged to generate the plots shown in the supplemental figures.

Reinforcement learning model

Request a detailed protocol

An adapted Q-Learning Reinforcement Model with five basic parameters was fit to the behavioral data produced by the relative reward serial reversal task (Daw, 2011; Sutton and Barto, 1998). Mouse choice patterns and outcome history were the primary inputs of the model. In order to capture trait-like characteristics of mouse behavior, behavioral sessions from the high and low reinforcement rate environments (four sessions each) were grouped and entered into the model together. The values of the lateralized choice alternatives were initiated at 0 and updated as follows:

Qt+1=Qt+α(RtQt),where
Rt=Vtγ

In this model, Qt is the value of the action taken on trial t and Rt is the function that approximates the perceived reward volume resulting from that action. Rt is defined as a compressive transformation of the reward volume, Vt, delivered after a choice raised to the coefficient, γ. γ is the compression parameter that relates how sensitively mice respond to reward volumes of different magnitudes. RtQt then, represents the reward prediction error (RPE) – the discrepancy between expected and realized reward – on trial t. The RPE is scaled by the learning rate (α), which determines the extent to which new information about the state-action pairing alters subsequent behavior. The scaled RPE is then used to update the value of the chosen action for the subsequent trial t+1. The value of the unchosen alternative was not altered on any trial and did not decay.

We utilized a modified softmax decision function to relate calculated action values with choice probabilities. The probability of choosing an alternative A on trial t was defined as:

PA(t)=11+ez,where
PA(t)= 11+ ez  , 

The inverse temperature parameter, β, is the conversion factor linking theoretical option values with realized choice output. High values of β indicate a tendency to exploit differences in action values, while lower values suggest more exploratory behavior. QA(t)QB(t) is the value of alternative A relative to the value of alternative B. In order to compare β across animals, this relative difference is scaled by 12γ, representing the maximum Q value (as largest delivered reward was 12 µL). To account for the influence of proximal choice output on subsequent decisions, we included the parameter κ – the persistence factor. This measure captures the extent to which the animal’s choice on the t−1 trial influences its choice on the t trial irrespective of outcome. Ct1 is an indicator variable that denotes whether the animal selected alternative A on the previous trial (Ct1=1) or if it selected alternative B (Ct1=1). To account for potential differences in bias between sessions, a bias term, cx, with an indicator variable Envx, was added for each session that the animal performed. This constant term captures spatial biases that animals have or develop in the course of a behavioral session. We performed a maximum likelihood fit using function minimization routines of the negative log likelihood of models comprised of different combinations of our three parameters (α, β, γ, κ, c) in MATLAB (Vo et al., 2014). In order to resolve global minima, the model was initiated from 75 random initiation points in the parameter space.

Fiber photometry

Request a detailed protocol

Viral injection and fiberoptic cannula implantation

Request a detailed protocol

Trained Nex+/- Nrxn1α-/- (n = 8) and Nex+/- Nrxn1αc/c (n = 6) mice were injected with adeno-associated viruses and implanted with a custom fiberoptic cannula on a stereotaxic frame (Kopf Instruments, Model 1900). Anesthesia was induced with 3% isoflurane + oxygen at 1 L/min and maintained at 1.5–2% isoflurane + oxygen at 1 L/min. The body temperature of mice was maintained at a constant 30°C by a closed loop homeothermic system responsive to acute changes in internal temperature measured via rectal probe (Harvard Apparatus, #50–722F). After mice were secured to the stereotaxic frame, the skull was exposed and anatomical landmarks bregma and lambda were identified. The skulls of the mice were subsequently leveled (i.e. bregma and lambda in the same horizontal plane) and 0.5 mm holes were drilled on regions of the skulls above the target locations. A pulled glass injection need was used to inject 300 nL of retroAAV2.EF1α−3xFLAG-Cre into the substantia nigra reticulata (SNr; AP: −4.2 mm, ML: +/−1.25 mm, DV: −3.11 mm) followed by 500 nL of AAV5.hSyn-DIO-GCamp6f into the dorsomedial striatum (DMS: AP: 0.85 mm, ML: +/−1.35 mm, DV: −2.85 mm). Holes were drilled ipsilaterally and injections were performed unilaterally per mouse. Virus was infused at 125 nL/min using a microinfusion pump (Harvard Apparatus, #70–3007) and injection needles were left in position for 10–20 min to allow diffusion of the viral bolus.

To implant each fiber optic, two 0.7 mm bore holes were drilled ~2 mm from the DMS skull hole. Two small screws were secured to the skull in these bore holes. A 400 μm fiberoptic cannula was lowered into the DMS injection site. Small abrasions on the skull surface were created with a scalpel, following which, we applied dental cement (Den-Mat, Geristore A and B) to secure the fiber optic placement. After surgery, mice were given oxygen at 2 L/min to aid in regaining consciousness. Mice were incubated for 4–6 weeks before recordings were performed. Approximately 2 weeks post-op mice were food deprived and reintroduced to the serial reversal task previously described. All data for photometry was collected only from 12 µL versus 0 µL sessions.

Data acquisition

Request a detailed protocol

Before recording sessions, mice were attached to a fiber-optic patch cord (400 μm core, 0.48 NA; Doric Lenses) to enable recordings. Patch-cords were attached to a Doric 4-port minicube (FMC4, Doric Lenses) to regulate incoming and outgoing light from the brain. An LED light driver (Thor Labs, Model DC4104) delivered alternating blue (470 nm, GCamp6f excitation) and violet (405 nm, autofluorescence/movement artifact) light to the brain. Light was delivered at ~50μW. The resulting excitation emissions were transferred through a dichroic mirror, a 500–550 nm filter, and were ultimately detected by a femotwatt silicon photoreceiver (Newport, Model 2151).

After attachment to the fiber-optic, animals were given a 5 min window to recover from handling before the initiation of a session. All recorded mice were trained to perform the relative reward serial reversal task before surgery. Animals were reintroduced to the task ~2 weeks post-surgery. At 3 weeks, expression of the GCamp6f construct was assessed and animals were trained to perform the task with the attached fiber-optic. After a minimum of 4 weeks and three full training sessions with the fiber optic, animals were eligible for recordings. Sessions lasted 1 hr. We introduced a 0–1 temporal jitter after the ITI and before the choice period to aid in dissociating task events.

Signal processing and analysis

Request a detailed protocol

Raw analog signals from behaving mice were demodulated (Tucker Davis Technologies, RZ5 processor) and recorded (Tucker Davis Technologies, Synapse). Demodulated 470 nm and 405 nm signals were processed and analyzed using custom Matlab (MathWorks, R2018b) scripts that are freely available upon request. Signal streams were passed through the filtfilt function, a zero-phase digital filter that filters data in both the forward and reverse direction to ensure zero phase distortion. Next, the data were down-sampled to 20 Hz. To account for bleaching of background autofluorescence in the patch cords over long recording sessions, the demodulated 470 nm and 405 nm signals were baselined to zero (the last value in the recording was used as an offset to have the signal decay to 0) and were fitted with cubic polynomial curves, which were subsequently subtracted from the signals. The ΔF/F of the debleached signals was calculated by sorting values into a histogram (100 bins) and then selecting the largest bin as the baseline signal. This baseline was subtracted from the raw 470 and 405 and then those values were divided by the baseline (note that the operation below was performed on both 470 and 405) [ΔF/F = (debleach(a)−baseline)/baseline]. Following this, the 405 nm control signal was subtracted from the 470 nm GCamp6f emission signal. The subtracted ΔF/F was transformed into z-scores by subtracting the mean and dividing by the standard deviation of a 2 min window centered on each point (1 min in front and behind). These standardized fluorescence signals were used for all subsequent analysis and visualization. The Bpod State Machine delivered electronic TTLs marking behavioral events to Synapse Software, which recorded their time and direction.

Modeling signal dynamics

Request a detailed protocol

The dynamics of preinitiation signal components was modeled as function of action output in the form of upcoming choice behavior (choice lateralization relative to implant [Choice], stay/shift behavior [Stay], explore/exploit behavior[Explore]), reward (reward volume on previous trial [RewardHist], reward prediction error [RPE] on previous trial and the relative action value on the current trial [ΔQ*temp−1]), prior signal dynamics (the preinitiation slope and integral on the previous trial [PIS and PIT], respectively), and the latency to initiate trials [LatInit]. Because the slope occurs after the integral on every trial and because slope and integral components are anti-correlated, the preinitiation integral on the t trial was included as a regressor in the modeling of the slope component. To account for individual animal differences in preinitiation signal components, we utilized a linear mixed-model:

PreinitiationIntegral RewardHist+RPE+Qtemp1+Choice+Stay+Explore+PIT+LatInit+(1|Subject)+1
PreinitiationSlope RewardHist+RPE+Qtemp1+Choice+Stay+Explore+PIS+LatInit+PreInitiationIntegral+(1|Subject)+1

Histology and immunohistochemistry

Request a detailed protocol

Mice were perfused via the left ventricle of the heart with 10 mL of 90% formalin. Whole brains were isolated and post-fixed in formalin overnight; 50 µm coronal and sagittal slices were sectioned in PBS. Slices from mice included in behavioral experiments were immediately mounted on microscope slides for imaging on an automated fluorescence microscope (Olympus BX63) at 10× (Olympus, 0.4NA). Additional sections were blocked in 3% normal goat serum in PBS for 1 hr and incubated with primary antibody overnight (1:500 Chick anti-GFP, abcam 13970; 1:1000 Mouse anti-FLAG, Sigma F1804). The following day, slices were washed with PBS and incubated for 3 hr with secondary antibody (1:1000 Goat Alexa488-conjugated anti-Chick, abcam 150173; 1: 1000 Goat Alexa647-conjugated anti-Mouse, Invitrogen #A-21235). Slices were washed 3× with PBS for 30 min and mounted on slides. Images were acquired from the same epi-fluorescent microscope as other images.

Electrophysiology

Mice were deeply anesthetized and perfused transcardially with ice-cold ACSF containing (in mM): 124 NaCl, 2.5 KCl, 1.2 NaH2PO4, 24 NaHCO3, 5 HEPES, 12.5 glucose, 1.3 MgSO4, 7H2O, 2.5 CaCl2. The brain was rapidly removed and coronal sections (250 μM thickness) were cut on a vibratome (VT1200s, Leica) in ice-cold ACSF. Sections were subsequently incubated <15 min in a NMDG-based recovery solution containing 92 NMDG, 2.5 KCl, 1.2 NaH2PO4, 30 NaHCO3, 20 HEPES, 25 glucose, 5 sodium ascorbate, 2 thiourea, 3 sodium pyruvate, 10 MgSO4, 7H2O, 0.5 CaCl2. The identity of retrogradely infected SPNs was visualized through viral fluorescence. Whole-cell recordings for mIPSCs were made using an internal solution containing (in mM): 135 CsCl, 10 HEPES, 0.6 EGTA, 2.5 MgCl, 10 Na-Phosphocreatine, 4 Na-ATP, 0.3 Na-GTP, 0.1 Spermine, 1 QX-314. mEPSCs were recorded using an internal solution containing (in mM): 115 CsMeSO3, 20 CsCl, 10 HEPES, 0.6 EGTA, 2.5 MgCl, 10 Na-Phosphocreatine, 4 Na-ATP, 4 Na-GTP, 0.1 Spermine, 1 QX-314 (pH adjusted to 7.3-7.4 with CsOH). Miniature spontaneous events were recorded in the presence of Tetrodotoxin (TTX; 1 μM), 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide (NBQX; 10 μM), D-(-)−2-amino-5-phosphonopentanoic acid (D-APV; 30 μM) for mIPSCs, and TTX plus picrotoxin (100 μM) for mEPSCs. Electrophysiology data was acquired using custom-built Recording Artist software (Rick Gerkin), Igor Pro6 (Wavemetrics), and analyzed using Minianalysis (Synaptosoft).

Statistical methodology

Power analysis was conducted in G*Power 3.1.9.4 (Faul et al., 2007) to obtain the appropriate sample size for the comparison of relative reward stay values of Neurexin1α wild-type and mutant animals. A power analysis for repeated measures ANOVA with two groups (wild-type, mutant) and eight measurements (two reward probabilities, four relative reward ratios), at power of 0.80, an alpha level of 0.05, and a medium-large effect size (f = 0.40), indicated a required sample size of 12. The sample size, n, for each experiment is clearly labeled on figures and in figure legends. Animals were tested in a repeated design aimed to assess their reward sensitivity in various reward conditions. However, each reward condition was only recorded once per animal. Replicate information for RNA experiments can be found in the methods section of the manuscript. Criteria for exclusion are detailed in the methods section as well.

All data were initially tested with appropriate repeated measure ANOVA (Prism8.0). Univariate regressions were performed in Prism8.0. Multivariate linear regressions were performed using the fitlm function in MATLAB. Multivariate linear mixed models were performed using the fitlme function in MATLAB. Main effect and interaction terms are described within figures, figure legends, and the results. Preinitiation slope coefficients were calculated using the polyfit function in MATLAB. The integral of photometry signals was calculated using the trapz function in MATLAB.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
    Trial-by-Trial Data Analysis Using Computational Models
    1. ND Daw
    (2011)
    Oxford University Press.
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
    Goal-directed learning and obsessive–compulsive disorder
    1. CM Gillan
    2. TW Robbins
    (2014)
    Philosophical Transactions of the Royal Society B: Biological Sciences 369:20130475.
    https://doi.org/10.1098/rstb.2013.0475
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
    R: A language and environment for statistical computing, version 2.6.2
    1. R Development Core Team
    (2017)
    R Foundation for Statistical Computing, Vienna, Austria.
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
    Reinforcement Learning: An Introduction
    1. RS Sutton
    2. AG Barto
    (1998)
    The MIT Press.
  82. 82
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89
  90. 90
  91. 91
  92. 92
  93. 93
  94. 94
  95. 95

Decision letter

  1. Mary Kay Lobo
    Reviewing Editor; University of Maryland, United States
  2. Michael J Frank
    Senior Editor; Brown University, United States
  3. Talia Lerner
    Reviewer; Northwestern University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

Behaviors that are oriented toward obtaining a specific goal are essential for normal life decision making but are impaired in neuropsychiatric disorders but the precise molecules that contribute to disrupted goal directed behaviors are unclear. This study demonstrates that removing Neurexin 1, a molecule that acts at the synapse to connect neurons, in excitatory brain circuits can disrupt goal directed behaviors and neural signals relating to reinforcement of these behaviors. These studies provide new information into the molecules that may underlie disrupted decision making processes in neuropsychiatric disorders.

Decision letter after peer review:

Thank you for submitting your article "Disruption of Nrxn1a within excitatory forebrain circuits drives value-based dysfunction" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Michael Frank as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Talia Lerner (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This study examines the role of Neurexin1a, a neuropsychiatric disorder-associated protein, in value-based decision making in mice. The authors address this by using a Neurexin 1a knockout and demonstrate deficits in optimizing selection strategies in an operant task with variable reward outcomes in this mouse line. Similar behavioral deficits are observed with a Neurexin1a conditional knockout (cKO) crossed to the Nex1-Cre, a strain that is supposed to express Cre-recombinase in postmitotic progenitors of forebrain projection neurons. Further, using fiber photometry paired with genetically encoded calcium sensors the authors show that Nex-Cre-Neurexin1a-cKO mice display deficits in activity of striatal direct pathway spiny projection neurons (dSPNs) prior to choice selection. Overall this study characterizes a genetic contribution to circuit dysfunction for neuropsychiatric disease relevant behaviors, pertaining to how animals choose actions according to cost and benefit.

Essential revisions:

1) There are concerns with the specificity of the genetic approach used. It appears that the conditional genetic approach was used to allow for developmental knockout of Nrxn1a as opposed to deletion in the adult animal. Can the author's please clarify and justify the need for using the developmental knockout approach? Further, addressing the developmental role of Nrxn1a and the human literature if applicable could improve the interpretations of the genetic approach used. Additionally, the Nex-Cre mice that are used for deletion of Nrxn1a from forebrain projections, has also been reported in mid- and hindbrain structures. Thus, it is difficult to interpret if behavioral abnormalities in these mice are due to deletion of Nrnx1a from forebrain projections or if they arise due to deletion of Nrxn1 in other regions of the brain. There is also misleading use of "cortical" rather than "forebrain" in the manuscript. Overall, clarification in text edits would improve the rationale for this genetic approach and clarify the specificity of the Nex-Cre approach.

2) There are concerns with the fiber photometry data that should be addressed. The z score values are low. In peak detection algorithms typically thresholds for something being defined as a significant peak at least 2.0 standard deviations above the median or mean average deviation of the data (some studies even use a cutoff of 2.9). However, here the Zscores are 0.2. It is difficult to determine if this is due to how Zscores were calculated or if there are not actually significant events detected. To address this please plot control data – either with a fluorophore (eYFP) or an isosbestic control (although eYFP is preferable) to ensure that fluctuations are due to detected calcium and not just pH shifts or movement artifacts that are occurring during the task.

Additional concerns are the difference in photometry signals between Nex-control and Nex-cKO animals in Figure 8, which could be due to the additional injection of cre (retroAAV2.EF1α-3xFLAG-Cre) in Nex-Nrxn1a cKO animals. This could potentially affect neurexin expression and activity of the direct pathway SPNs. Another control (Nrxnc/c; Nex+/+) would help to resolve this concern.

3) There is misleading use of "reward-associated neural signals" and "value-related neural signals" in reference to dSPN activity prior to the initiation of a trial. While the activity is interesting it does not relate to reward presentation or consumption. Please also show dSPN activity at the time of choice and reward consumption in addition to trial initiation, as the activity at time of choice relates to decision making and activity at the time of reward relates to value-encoding.

4) There are concerns that the Neurexin1a-KO mice have a working memory deficit. The t-1 regression coefficient is lower and other coefficients aren't higher, suggesting that the KOs are not using as much information from past trials to guide their actions as controls are. Furthermore, the failure to modulate initiation latency based on previous outcome could be because they don't remember it easily. Finally, the differences between the control and KO mice are more apparent with lower probability of reward, perhaps because it's harder for them to remember with so much uncertainty. Two experiments could help control for this: (1) test working memory directly e.g. in a spontaneous alternation task, and (2) run a version of the authors' task with P=1. When P=1, uncertainty is not an issue, but the representation of two choices and their values is.

5) There are concerns with the interpretation of deficits in responses to relative reward outcomes in the Neurexin1a-KO. In Figure 1, Neurexin1a-KO animals show deficits in responses to relative reward outcomes. Specifically, the largest deficit observed was at the peak of the deltaSucrose. One interpretation is that Neurexin1a-KO animals simply consume less sucrose than WT's and therefore are less motivated to select the higher sucrose volumes, not because there is a deficit in optimizing operant strategies. Thus additional experiments examining whether Neurexin1a-KO mice consume comparable amounts of sucrose as WT animals in a standard operant paradigm or in a free-access model would help to address this concern.

6) There is a disconnect between the behavioral results in the Nex-Neurexin1a-cKO animals (which the authors state are targeted to forebrain neurons) and subsequent neural imaging studies targeted to direct pathway neurons of the dorsal striatum. While the authors discuss the importance of the dorsal striatum in reward-based tasks, they fail to address how Neurexin1a deficits in the prefrontal cortex/forebrain would lead to deficits in dorsal striatum activity during the task. Please provide clarification in the text.

7) Many of the findings remain largely correlative in nature. While the reviewers do not see a need for further experimentation to show a gain of function the authors should carefully examine some conclusions to indicate the correlative but non-causal results.

8) Some of the figures should be edited for clarity. For example, in some figures such as 1B, the statistical differences are entirely unclear. Although there is a described overall genotype difference, it is unclear if there was an interaction and at what reward size these differences were statistically important. This lack of clarity can be seen in most of the figures. Further, based on the information presented here, it is not possible to determine interactions or how post-hoc analysis was completed. While the figure legends contain some statistical information, they do not lead the reader to fully understand the important changes. Please also include how any post-hoc comparisons were done.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "Disruption of Nrxn1a within specific excitatory forebrain circuits drives value-based dysfunction" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Michael Frank as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: David Dietz (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary:

This study examines the role of Neurexin1a, a neuropsychiatric disorder-associated protein, in value-based decision making in mice. The authors address this by using a Neurexin 1a knockout and demonstrate deficits in optimizing selection strategies in an operant task with variable reward outcomes in this mouse line. Similar behavioral deficits are observed with a Neurexin1a conditional knockout (cKO) crossed to the Nex1-Cre, a strain that is supposed to express Cre-recombinase in postmitotic progenitors of forebrain projection neurons. Further, using fiber photometry paired with genetically encoded calcium sensors the authors show that Nex-Cre-Neurexin1a-cKO mice display deficits in activity of striatal direct pathway spiny projection neurons (dSPNs) prior to choice selection. Overall this study characterizes a genetic contribution to circuit dysfunction for neuropsychiatric disease relevant behaviors, pertaining to how animals choose actions according to cost and benefit.

Essential revisions:

While the revisions have improved the study there are additional revisions needed. Below the remaining concerns are listed including how each should be addressed by tempering the language, re-analysis of data, or the inclusion of one new additional experiment.

1) Please address the below comment and comment #2 by performing re-analysis with the collected data and providing additional details. New experiments are not required.

Even with the additional control data provided, the fiber photometry data, as presented, is difficult to interpret. The Z-score peaks presented in Figure 7D-F (0.2 Z scores) are significantly lower than peaks seen in the overall representative trace in Figure 7C (where there are peaks of up to ~4 Z-scores). While the authors suggest that the small z-scores are due to genetically defined populations, work from several other groups recording from comparable cell populations have seen much larger task-specific peaks in GCaMP-measured activity (Cui et al., 2013 Nature; Cui et al., 2014 Nat. Prot.; Klaus et al., 2017 Neuron; Calipari et al., 2016 PNAS), demonstrating that this alone cannot explain this. There are several levels of data analysis that are missing from the manuscript that would alleviate concerns regarding this issue:

a) Changes in Ca2+ signaling should be presented in a trial-by-trial basis in addition to the summary seen in Figure 7.

b) Trial-specific traces should be presented in a non-Z-scored fashion (see comment 2 for why this is particularly important).

c) Although the data from the 405 control channel is presented, it is presented as an internally controlled Z-score. The 405 traces should be plotted at the same scale as the 470 trace to preclude the signal detected being movement artifacts. There is no mention of scaling in the analysis pipeline and this is critical to using the 405 as a control.

d) The authors report the laser power at 50μW, which is very low for both recording channels – especially the 405 – and could in part explain the flat line for the 405 channel. i.e. the light power was not high enough to detect events in that channel.

e) In the Materials and methods the authors mention that the data is filtered but provide no additional information about the filter. What is the filter, what is the equation, how does this filter alter the timing of the signal relative to the events? This is critical to understand the analysis pipeline.

2) The biggest issue in the data processing is with how the z-scores were calculated for the peri-event records. Because the signals were done as a change from the entire peri-event trace this makes it so that the median of the trace is 0. For signals with bigger peri-event records this will just move the pre and post response traces down. It will look like the peaks are similar, but the ramp up and down are different. This would then make the data look like the complete opposite of the raw data. The reason this is concerning here is that they data for the largest event shows this pattern, where the baseline is lower, rather than the event. Nearly all of the work on striatal populations has shown that these circuits scale with reward value – with larger peak response around the larger outcome. Thus, these data may actually show the opposite findings when plotted in a more appropriate way.

3) Please address the below concern by tempering the language in the manuscript.

Although the authors address concerns regarding specificity of the NEX-Cre/Nrx1aKO mouse line and potential mechanistic link between forebrain function in the text, there remains a large disconnect between the widespread loss of neurexin1a expression and the specific effects seen in the DMS->SNr circuitry. As the authors mention, the NEX-Cre has primary recombination sites in the amygdala, hippocampus, and several PFC subregions; each which feed into the DMS-> SNr pathway and each provide a unique aspect to reward-associated behavior. A key premise of the manuscript is that loss of forebrain neurexin1a leads to disruption of the DMS->SNr circuitry during reward-seeking. However, what is missing from the manuscript is how is this circuit disruption occurring. Given the widespread loss of neurexin1a expression, it is difficult to infer a potential mechanism or pathway as a source for DMS activity disruption, beyond likely glutamatergic input.

4) Please address the below concerns by performing an additional experiment to examine mEPSCs.

A main lingering concern is still about the retroAAV cre injection into SNr. The mIPSC recordings in DMS show that one particular change (in inhibitory transmission) does not occur but doesn't rule out other possibilities, particularly in terms of potential changes in excitatory transmission. It is assumed that the mEPSCs in Nex-cre Nrxn1a-cKO mice change in comparison to Nex-cre controls (that is their hypothesis, that corticostriatal deficits occur and underlie the changes in dSPN activity), although they don't show it but reference unpublished data in the reviewer response. A retroAAV cre injection in Nrxn1a cKO-only animals (no Nex-cre) to determine that mEPSCs don't change if the Nex-cre isn't there would be important. This would show that retroAAV cre expression doesn't do anything additional that is not intended at their synapse of interest.

More generally, injection of the retroAAV cre virus into Nrxn1a-cKO mice in SNr would cause recombination in SNr as well as in all input structures of SNr (not only DMS). The basal ganglia are so interconnected this could cause strange unforeseen consequences for dSPN activity.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "Disruption of Nrxn1a within excitatory forebrain circuits drives value-based dysfunction" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Michael Frank as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: David Dietz (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to accept without delay manuscripts, like yours, that they judge can stand as eLife papers without additional data, even if they feel that they would make the manuscript stronger. Thus the revisions requested below only address clarity and presentation.

Summary of the work:

Behaviors that are oriented toward obtaining a specific goal are essential for normal life decision making but are impaired in neuropsychiatric disorders. The precise molecules that contribute to disrupted goal directed behaviors are unclear. This study demonstrates that removing Neurexin 1, a molecule that acts at the synapse to connect neurons, in excitatory brain circuits can disrupt goal directed behaviors and neural signals relating to reinforcement of these behaviors. These studies provide new information into the molecules that may underlie disrupted decision making processes in neuropsychiatric disorders.

Recommendations:

It is suggested that the data presented in the response to reviewers be accessible so that the readers can see the raw data minimally processed. A concern hinges upon data that is included in the response where the z-scored data is introducing changes into the data that is not seen in the raw df/f data. Author response image 5 clearly shows that there are significant differences between A – unprocessed data and B- z-scored data that makes it look like huge decreases are happening that do not seem apparent in the unprocessed data. While this could be fine if the authors are transparent with their analysis pipeline, providing the data would help the reader draw their own conclusions.

It is recommended that the authors include this data or make it accessible to the readers in some form (i.e. link, website, etc).

https://doi.org/10.7554/eLife.54838.sa1

Author response

Essential revisions:

1) There are concerns with the specificity of the genetic approach used. It appears that the conditional genetic approach was used to allow for developmental knockout of Nrxn1a as opposed to deletion in the adult animal. Can the author's please clarify and justify the need for using the developmental knockout approach? Further, addressing the developmental role of Nrxn1a and the human literature if applicable could improve the interpretations of the genetic approach used.

Given Nrxn1a’s early expression around the onset of synaptogenesis, its role in initial synaptic maintenance and our desire to specifically relate our behaviors to the constitutive LOF allele, we chose gene removal early in development. While we agree that testing the continued necessity of Nrxn1a in adult structures is an interesting question, we believe it goes beyond our initial question of modeling Nrxn1a-associated brain dysfunction, which involves gene loss from early post-neurogenic periods. These specific adult maintenance functions will be pursued in a separate study. We have added text discussing the early expression of Nrxns prior to synaptogenesis and our desire to test this early function given its relevance to disease models.

Additionally, the Nex-Cre mice that are used for deletion of Nrxn1a from forebrain projections, has also been reported in mid- and hindbrain structures. Thus, it is difficult to interpret if behavioral abnormalities in these mice are due to deletion of Nrnx1a from forebrain projections or if they arise due to deletion of Nrxn1 in other regions of the brain. There is also misleading use of "cortical" rather than "forebrain" in the manuscript. Overall, clarification in text edits would improve the rationale for this genetic approach and clarify the specificity of the Nex-Cre approach.

Regarding the specificity of the Nex-Cre allele, our own in-lab analysis together with analysis of the original paper (Goebbels et al., 2006) suggest that expression in mid and hindbrain structures is minimal. While Table 1 in the original publication lists an enormous number of sites of recombination, it makes no distinction between sparse recombination (observed in mid/hindbrain, see Figure 3, Goebbels et al., 2006) and the primary recombination sites (cortex, hippocampus and amygdala). Recent publications (Kramer et al., 2018 and Bimpisidis et al., 2018) have highlighted a small (~10%) subpopulation of mVTA neurons that are NEX positive and captured by the Nex-Cre allele. While we cannot formally rule out a contribution of these neurons to our observed behavior, they are unlikely to contribute to our Ca2+ imaging findings, as this NEX-Cre DA subpopulation has been shown to project exclusively to the medial shell of the NAc (see Figure 3B, Kramer et al., 2018). Nex-negative DA neurons make up the entire population projecting to the DMS, where all of our recordings were performed. We have added a discussion of these caveats in our Discussion section.

Finally, we agree that our terminology needs to be clarified for publication. Beyond the confusion created by our interchanging “cortical” and “forebrain,” we realized that “forebrain excitatory neurons” describes both cortical, hippocampal, amygdala AND thalamic projections. We have decided on using “telencephalic excitatory” as this includes the neocortical, hippocampal and subset of amygdalar neurons captured by the Nex-Cre allele.

2) There are concerns with the fiber photometry data that should be addressed. The z score values are low. In peak detection algorithms typically thresholds for something being defined as a significant peak at least 2.0 standard deviations above the median or mean average deviation of the data (some studies even use a cutoff of 2.9). However, here the Zscores are 0.2. It is difficult to determine if this is due to how Zscores were calculated or if there are not actually significant events detected. To address this please plot control data – either with a fluorophore (eYFP) or an isosbestic control (although eYFP is preferable) to ensure that fluctuations are due to detected calcium and not just pH shifts or movement artifacts that are occurring during the task.

Z scores were calculated as the signal-mean/standard deviation of the signal over a moving 60 sec. window, to minimize the effect of broader signal fluctuations. The signals shown in Figure 7/8 and those used for correlation analysis do not employ peak detection algorithms, but instead peri-event averaged signals. The small signal fluctuations seen here likely result from the sparseness of the retrograde labeling employed to distinguish dSPNs and the relatively small motor movements required in our BPod behavioral machines, where mice can register choices with minimal motor output. Here we provide data on our 405 isosbestic channel demonstrating the absence of similar waveforms on this Ca2+ independent channel, highlighting the specificity of our 470nm signal. We have added this analysis in Figure 7—figure supplement 1A which accompanies main Figure 7, and a reference in the text.

Furthermore, we include in Author response image 1 examples of Ca2+ waveforms acquired in Med Associates boxes with a non-retrograde approach to label dSPNs (direct DIO-GCamp6f into D1Cre transgenic). While these averaged waveforms exhibit larger z-scored signals (note changed y-axis), they have very similar waveform shapes in the period leading up to initiation (compare gray shaded boxes, which bracket the same time window around task initiation).

Author response image 1

Finally, we opted to analyze total signal waveforms as opposed to peaks because (1) it involved less transformation of the original photometry data; (2) the unclear significance of event-associated peaks; (3) our stringent criteria for peak detection lead to the omission of many trials in which initiation-relevant peaks were undetected. Nevertheless, to address reviewer concerns, we have performed a similar analysis using an event-associated peak detection algorithm focusing on a 1sec. window prior to initiation. To check our key results with this alternative analysis, we examined event-associated Ca2+ peak amplitude correlations with trial-by-trial behavioral parameters as in Figure 8. Consistent with our prior analysis, we found that multiple behavioral variables of value processing exhibit significant correlations with pre-initiation event-detected peaks and that these correlations are absent in Nex cKOs (see Author response image 2). Given the smaller sampling of data from event-associated peak-detection analyses and greater confidence in our waveform data, we opted to omit these data from the final manuscript. Nevertheless, the main take-home point that value-correlated neural signals are abrogated in Nex cKOs stands regardless of analysis method used.

Author response image 2

Additional concerns are the difference in photometry signals between Nex-control and Nex-cKO animals in Figure 8, which could be due to the additional injection of cre (retroAAV2.EF1α-3xFLAG-Cre) in Nex-Nrxn1a cKO animals. This could potentially affect neurexin expression and activity of the direct pathway SPNs. Another control (Nrxnc/c; Nex+/+) would help to resolve this concern.

We acknowledge this potential caveat which resulted from a desire to specifically label one subtype of striatal SPN, the existing tools (no robust retroAAV2-FLP virus), and concern at the time for behavioral phenotypes in the Nex-Cre line, which had never been used in these types of behaviors before. Nevertheless, we do not believe that this design is the likely cause of our results for the following reasons: (1) the differing timelines for gene removal between the Nex-mediated recombination (early embryonic) as compared to rAAV2-Cre (2-3 weeks before behavior); (2) Nrxn1a expression levels in striatal SPNs are several orders of magnitude lower than that in cortex (Fuccillo et al., 2015) making LOF effects less likely; (3) given Nrxn’s exclusive presynaptic function, the fact that retrograde Cre will cause deletion in the recorded SPNs is unlikely to alter excitatory synaptic transmission that largely drives recorded Ca2+ signals in this population; (4) while the sparse nature of the retrograde labeling is unlikely to impact a large enough number of dSPNs to significantly alter their lateral inhibition, we directly tested this possibility by injecting rAAV2-3xFlag-Cre or rAAV2-EGFP-∆Cre (an enzymatically inactive truncated version of Cre) into Nrxn1aC/C; NexCre adults, waiting the same duration of time as for our behavior/recording analysis (2-3 weeks) and then recorded mIPSCs in acute slices cut from these mice. We did not note any significant change in the frequency or amplitude of mIPSCs in neurons with Cre or δ-Cre expression, suggesting alterations in inhibition are unlikely to account for our changes in recorded Ca2+ signal. Taken together, we believe the most straight-forward explanation of our data is that deletion of Nrx1a from cortical progenitors alters the excitatory synapses of these projections pathways, with one result being the abrogation of value-related signals in a key target, the DMS. We have included this data in a new Figure 8—figure supplement1 (accompanies Figure 8) Furthermore, we have added a discussion of this experimental caveat to our revised Results.

3) There is misleading use of "reward-associated neural signals" and "value-related neural signals" in reference to dSPN activity prior to the initiation of a trial. While the activity is interesting it does not relate to reward presentation or consumption. Please also show dSPN activity at the time of choice and reward consumption in addition to trial initiation, as the activity at time of choice relates to decision making and activity at the time of reward relates to value-encoding.

We agree that the activity presented does not relate to reward presentation or consumption, as it is leading up to the initiation of the trial. We focused on this epoch for several reasons: (1) it is the most clearly temporally dissociated signal, as compared with the choice and reward-retrieval signals which have shorter and less variable inter-event latencies (200-400msec), making them challenging to parse using GCamp-based Ca2+ indicators; (2) we have multiple lines of unpublished evidence (in two different behavioral settings) that these pre-initiation signals are strongly modulated by the upcoming/current value of trials. While we agree that choice-associated signals might also be relevant, we do not think this invalidates our decision to focus on the initiation period, as this is the part of the behavior that demonstrates the greatest variability according to prior outcomes, with the choice and reward retrieval being more sterotyped actions resulting from extensive training in this specific paradigm. For revision, we have edited out the use of “reward-associated neural signals,” instead using “value-modulated neural activity,” or “value-modulated initiation activity,” as appropriate. Furthermore, as requested, we have provided the choice/retrieval-associated activity as requested in Figure 7—figure supplement 1B-E. While the downward choice waveforms are completely separated in control and overlapping in Nex cKOs, these differences likely relate to the preceding signal differences in value-modulated initiation activity. To quantify the choice-related signal we fit the choice aligned waveforms with a linear model and compared the slope values (Figure 7—figure supplement 1D) for control and Nex cKO mice (Figure 7—figure supplement 1E). We did not observe modulation of this slope by prior outcome or across genotype, suggesting that p-dSPN population Ca2+ activity at choice is not strongly modulated by value.

4) There are concerns that the Neurexin1a-KO mice have a working memory deficit. The t-1 regression coefficient is lower and other coefficients aren't higher, suggesting that the KOs are not using as much information from past trials to guide their actions as controls are. Furthermore, the failure to modulate initiation latency based on previous outcome could be because they don't remember it easily. Finally, the differences between the control and KO mice are more apparent with lower probability of reward, perhaps because it's harder for them to remember with so much uncertainty. Two experiments could help control for this: (1) test working memory directly e.g. in a spontaneous alternation task, and (2) run a version of the authors' task with P=1. When P=1, uncertainty is not an issue, but the representation of two choices and their values is.

It is true that a deficit in working memory (specifically relevant here as to which choice was previously made) can generate phenotypes similar to abnormalities in value coding for a task in which choice is the only read-out of internal processing we have. However, we don’t believe our data support this conclusion for the following reasons: (1) the lower coefficient for the large reward from the logistic regression models reflects the diminished reinforcing property of the reward; (2) if the mice had working memory deficits, we would also expect differences in the choice regression coefficient, which are not observed; (3) if the deficits were driven by working memory deficits, we would expect uniform choice abnormalities across all reward contrasts, instead we observe trends in which the largest choice differences by genotype are for the largest differences in reward contrast. However, to more directly address the reviewer’s concern, we have performed spontaneous alternation as requested on our Nex cKOs and their littermate controls. Consistent with normal working memory function in Nex cKOs, we do not observe any genotypic differences. These data have now been included in modified Figure 5—figure supplement 1G and in the Results.

5) There are concerns with the interpretation of deficits in responses to relative reward outcomes in the Neurexin1a-KO. In Figure 1, Neurexin1a-KO animals show deficits in responses to relative reward outcomes. Specifically, the largest deficit observed was at the peak of the deltaSucrose. One interpretation is that Neurexin1a-KO animals simply consume less sucrose than WT's and therefore are less motivated to select the higher sucrose volumes, not because there is a deficit in optimizing operant strategies. Thus additional experiments examining whether Neurexin1a-KO mice consume comparable amounts of sucrose as WT animals in a standard operant paradigm or in a free-access model would help to address this concern.

While it is true that the largest choice deficits noted are at the peak of the reward contrast (see point #4 above), we do not believe this has to do with the Nrx1a mutants consuming less sucrose. As we show in additions to Figure 2—figure supplement 1, reward probability and reward contrast, but not genotype determine the total volume of reward consumed per session (Figure 2—figure supplement 1B). In addition, the latency to initiate trials over time (for pr75, 12v0) also suggests that knockout mice do not lose engagement with the task as it progresses (Figure 2—figure supplement 1C). We have added reference to these data.

6) There is a disconnect between the behavioral results in the Nex-Neurexin1a-cKO animals (which the authors state are targeted to forebrain neurons) and subsequent neural imaging studies targeted to direct pathway neurons of the dorsal striatum. While the authors discuss the importance of the dorsal striatum in reward-based tasks, they fail to address how Neurexin1a deficits in the prefrontal cortex/forebrain would lead to deficits in dorsal striatum activity during the task. Please provide clarification in the text.

Our thinking here was based on three key facts: (1) Neurexin1a is highly expressed in deep-layer cortical neurons that send projections to striatum; (2) In hippocampal synapses, Neurexin1a KOs exhibit impaired excitatory synaptic transmission; (3) unpublished data from the lab demonstrating excitatory synaptic deficits in corticostriatal circuits of Neurexin1a KO and heterozygote mutant mice. We have added this clarification to the text, both in the Results section and in the Discussion.

7) Many of the findings remain largely correlative in nature. While the reviewers do not see a need for further experimentation to show a gain of function the authors should carefully examine some conclusions to indicate the correlative but non-causal results.

We do not feel that the initial characterization of the constitutive KO mice (Figure 1-4), nor the cell type-specific perturbations (Figure 5,6) are in any manner correlative. In fact, circuit-specific genetic manipulations are rarely, if ever, analyzed in such complex behavioral tasks. We do acknowledge that the analysis of neural signals in the Nex-specific deletions relies heavily on associations between population Ca2+ signals and aspects of behavioral performance (Figure 7,8). Given the novelty of our discovery, we felt it was important to uncover potentially relevant sites of Nrx1a-altered value coding before exploring neural circuit manipulations to prove causality. These complex experiments are currently underway in follow-up work to this study. For now, the best we can do is clarify the distinctions between our causal and correlative data throughout the manuscript.

With regard to “showing gain of function,” it is unclear to us what that experiment would look like and what question it could answer. While a rescue experiment to re-express Nrx1a in adult KOs would have profound implications for treatment, the large size of the Nrxn1a cDNA has impeded our attempts to make a AAV-based viral vector suitable for widespread infection and reliable Nrxn1a expression. In this manuscript we focused on circuit-specific loss-of-function approaches so we could compare them to the constitutive Nrx1a KO mice, as we were interested in modeling the loss-of-function contributions of Nrxn1a to reward processing.

8) Some of the figures should be edited for clarity. For example, in some figures such as 1B, the statistical differences are entirely unclear. Although there is a described overall genotype difference, it is unclear if there was an interaction and at what reward size these differences were statistically important. This lack of clarity can be seen in most of the figures. Further, based on the information presented here, it is not possible to determine interactions or how post-hoc analysis was completed. While the figure legends contain some statistical information, they do not lead the reader to fully understand the important changes. Please also include how any post-hoc comparisons were done.

We have included tables with the outcomes of statistical tests to accompany each figure, for clarity. When developing the protocols tested in this paper, we were interested in exploring whether relative reward discrepancy or probability interacted with mouse genotype to produce a choice deficit in unique reward environments (we hypothesized, for instance, that Nrxn1a mutants would have more severe deficits with increased reward scarcity). Each animal repeated the same task, but with slight variations in reward discrepancy and probability. This repeated design was influenced by our previous paper (Alabi et al., 2019), in which we establish that mice exhibit trait-like rates of performance across multiple sessions in a similar task.

Upon completing our analysis, we found a suspected effect for relative reward discrepancy on performance and reward stay behaviors, but no effect for reward probability. While there was a significant effect for genotype overall, there was no interaction between genotype and the other factors of interest. In other words, the average difference in performance/relative reward-stay between Nrxn1a WT and KO animals did not vary with the discrepancy in reward volume or reward delivery rate. As such, we did not report individual post-hoc tests comparing control and mutant performance in individual reward regimes, as these differences would not be easily interpreted. Additionally, these tests would be conducted using data from individual sessions only, subjecting them to natural session-to-session variability in mouse choice behavior. The repeated measures design was meant to decrease the effects of this variability in mouse behavior and the statistical test that encapsulates that is the ANOVA itself.

One method to better aid clarity is to z-score performance/reward-stay behavior for the population in each reward regime. Condensing z-scores (comparative performance/reward-stay) produces clear differences:

Author response image 3

We originally considered showing the data in this manner but decided that too much information was lost in condensing the data down like this.[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Essential revisions:

While the revisions have improved the study there are additional revisions needed. Below the remaining concerns are listed including how each should be addressed by tempering the language, re-analysis of data, or the inclusion of one new additional experiment.

1) Please address the below comment and comment #2 by performing re-analysis with the collected data and providing additional details. New experiments are not required.

Even with the additional control data provided, the fiber photometry data, as presented, is difficult to interpret. The Z-score peaks presented in Figure 7D-F (0.2 Z scores) are significantly lower than peaks seen in the overall representative trace in Figure 7C (where there are peaks of up to ~4 Z-scores). While the authors suggest that the small z-scores are due to genetically defined populations, work from several other groups recording from comparable cell populations have seen much larger task-specific peaks in GCaMP-measured activity (Cui et al., 2013 Nature; Cui et al., 2014 Nat. Prot.; Klaus et al., 2017 Neuron; Calipari et al., 2016 PNAS), demonstrating that this alone cannot explain this. There are several levels of data analysis that are missing from the manuscript that would alleviate concerns regarding this issue:

We are eager to address these points of analysis so that our paper will be clearer to all readers. There are several points that must be first clarified:

a) The Z-score peaks presented in Figure 7D-F are not incongruous with the peaks seen in the overall representative trace in Figure 7C. We note that on our previous submission, Figure 7C does not have a y-axis, so that it is not clear what the absolute magnitude (ΔF/F) of individual peaks are. However, the reviewer correctly points out that there are peaks that are about 4 z-scores greater than the mean/median of the signal. If the mean/median of the signal in Figure 7C were to be less than zero, a peak with a large prominence (that is, with a large amplitude versus surrounding signal) might have a lower absolute magnitude than anticipated. This is the case with the signals that we analyze.

b) When mice are engaged in the task, the average signal (or baseline) is lower than when they are disengaged and running around the cage. Mean signals for animals in engaged and disengaged epochs of the task were calculated. Engaged periods were defined as any timepoint within a window of 5 seconds before an initiation event (to start a trial) and after a nosepoke exit (to end a trial). Epochs outside of this window were defined as periods of disengagement. Note, again, that this phenomenon is seen both in the experiments described in this manuscript, but also in separate experiments in another behavioral context. We note no significant differences in the average signal of control and mutant mice either in the engaged period or the unanalyzed disengaged period. (Analyzed by Two-Way Repeated Measures ANOVA: Engagement, p=0.0003; Genotype, p=0.166; Interaction, p=0.337). We now add this data to Figure7—figure supplement 1.

c) It is important to note that given the lower average signal in engaged periods, the magnitude of initiation peaks may not be as high as we expect. Instead, as suggested above, we might expect the prominence (height versus surroundings) of initiation peaks to be high. If we look at the fast-peak phase of the large reward initiation trace in Figure 7F, for instance, we can see that while the magnitude of the peak is roughly 0.4 z-scores, this occurs on top of an immediate background of -0.4 z-scores. This means that the dynamic range of initiation-associated peaks is actually 0.8-1 z-scores, on average, rather than the 0.2 z-scores quoted by the reviewer in reference to 7D. This falls within the dynamic range of striatal photometry experiments in the field (and the ones cited by the reviewer).

d) We employed a task with structured behavioral elements that mice had to complete in order to progress with the task. This task structure allowed us to perform analysis of the photometry signal by assessing peri-event striatal dynamics. Note that, for the purposes of this manuscript, we limited our analysis to the pre-initiation time epoch. We have annotated Figure 7C, to highlight a few points. There are 4 initiations shown in Figure 7C. They have been numbered in green, for consistency. A line representing mean value of this trace is also included to represent the baseline of this signal segment, which is, again, less than 0 during this period of engagement with the task.

  • First, note that the shape of the preinitiation in individual trials roughly approximates the shape of the averaged signals. That is, a downward ramp in activity that precedes a peak in activity, marked with a green timestamp. This holds true even in trial 2, where there is a downward slope followed by a small uptick in activity before the initiation.

  • If we carefully examine the preinitiation slow ramp of each initiation event, we observe that population activity actually drops to either the mean or to a value below the mean value for this trace (which again, is less than 0 to being with). As a result, the subsequent fast peaks associated with these initiations are only slightly greater than the mean value (trials 1,2,4) and in one case actually less than the mean (trial 3). This is one of many reasons that we independently quantified both the ramp and the peak – because both processes can be observed in individual traces and because the two processes together dictate peak magnitudes.

  • As expressed by the reviewer, we note that there are indeed several peaks in Figure 7C of ~4 z-scores. It is important to note however, that those peaks are not associated with initiation events temporally. We performed an analysis on a specific behavioral epoch and that analysis is consistent across figures as demonstrated above. Initiation peak magnitudes are on average ~0.3 z-scores, but the dynamic range of peaks is 0.8-1 z-score. The highest peak values observed in Figure 7C do not occur in phase within our temporal window. These peaks are much more closely associated with nosepoke exits (light blue), occurring either just before or after these events. We further note that, 1) these large peaks exist, even though the data is z-scored and we are recording from a genetically defined cell population and 2) these peaks are not the topic of analysis for this paper because they are not in phase within the temporal window of interest (i.e. they are not associated with initiation events)

e) Finally, we would further emphasize that we purposefully employ a small, but genetically defined, cell population in our photometry experiments. In our previous rebuttal response, we show that in a similar task, in mice in which all DMS D1 neurons are labeled, signal dynamics around initiation are roughly twice as large (1.4-2 scores). The papers cited do not use conditional retrograde labeling of cell populations, instead viral GCamp expression in genetic mouse lines.

a) Changes in Ca2+ signaling should be presented in a trial-by-trial basis in addition to the summary seen in Figure 7.

b) Trial-specific traces should be presented in a non-Z-scored fashion (see comment 2 for why this is particularly important).

Presented in Author response image 4 are individual GCamp6f traces of trials in a control animal, aligned to initiation, sorted by prior large reward (blue) and small reward (red) trials. The left column are the z-scored data used in the paper and the right column is the raw ΔF/F(%) traces. Note that the trends seen in the averaged z-scored data can be observed in the z-scored data at the individual trial level and also observed in the individual non-z-scored data at the individual trial level. In contrast to the concerns noted below by the reviewer, there is actually a greater dynamic range with z-scoring rather than a flattening of the signal. Finally, comparing the z-scored and non-z-scored data demonstrates the signal drift and shifting baselines normally seen without signal standardization. Averaging at this level produces the same waveform shape but with a larger standard error, as demonstrated in Author response image 5.

Author response image 4
Photometry traces by trial.

10 individual traces selected from trials following large reward (blue) and trials following small reward (red). Panel A and C show z-scored data while B and C show raw ΔF/F(%).

c) Although the data from the 405 control channel is presented, it is presented as an internally controlled Z-score. The 405 traces should be plotted at the same scale as the 470 trace to preclude the signal detected being movement artifacts. There is no mention of scaling in the analysis pipeline and this is critical to using the 405 as a control.

We thank the reviewer for catching a mistake in the labeling of our axis – the 405 data shown here is the actual ΔF/F(%), not a Z-score. We have now made these corrections. In addition, we will clarify our photometry pipeline in the Materials and methods as follows: (1) We did not z-score the 405 signal before incorporating it. The 405 was subtracted from the 470 before z-scoring occurred. We agree with the reviewer that it does not make sense to z-score the 405. (2) The controlFit method for calculating ΔF/F relies on scaling the 405 to the 470, then subtracting the 405 from the 470. We, however, employed another method for calculating ΔF/F. To account for a steady decrease in baseline fluorescence over prolonged sessions, the 405 and 470 were baselined to zero (the last value in the recording was used as an offset to have the signal decay to 0). Following this, the data were fit with a cubic polynomial curve, which was then subtracted from both signals (bleach detrending). Afterward, both signals were standardized by sorting values into a histogram (100 bins) and then selecting the largest bin as the baseline signal. This baseline was subtracted from the raw 470 and 405 and then those values were divided by the baseline (note that the operation below was performed on both 470 and 405). Following this, the control signal was subtracted from the GCamp6f signal [dF/F(a) = (debleach(a)-baseline/baseline]1.

d) The authors report the laser power at 50μW, which is very low for both recording channels – especially the 405 – and could in part explain the flat line for the 405 channel. i.e. the light power was not high enough to detect events in that channel.

We note that there are a wide range of light intensities seen in fiber photometry experiments across the literature and at eLife itself. eLife has published papers in which laser intensity at the tip of the patch cable was approximately 5 μW 2 and others where light intensities were about 200 μW 3. Our measurement of 50uW falls right into this range of previously published laser light intensities.

e) In the Materials and methods the authors mention that the data is filtered but provide no additional information about the filter. What is the filter, what is the equation, how does this filter alter the timing of the signal relative to the events? This is critical to understand the analysis pipeline.

We use the filtfilt function in Matlab to digitally filter our data. This is a zero-phase digital filter that filters data in both the forward and reverse direction, resulting in zero phase distortion. As a result, it will not change the relationship of the signal to individual behavioral events. This information has been added to the Materials and methods.

2) The biggest issue in the data processing is with how the z-scores were calculated for the peri-event records. Because the signals were done as a change from the entire peri-event trace this makes it so that the median of the trace is 0. For signals with bigger peri-event records this will just move the pre and post response traces down. It will look like the peaks are similar, but the ramp up and down are different. This would then make the data look like the complete opposite of the raw data. The reason this is concerning here is that they data for the largest event shows this pattern, where the baseline is lower, rather than the event. Nearly all of the work on striatal populations has shown that these circuits scale with reward value – with larger peak response around the larger outcome. Thus, these data may actually show the opposite findings when plotted in a more appropriate way.

To address these concerns, we will elaborate on our reasons for z-scoring and establish that this method does not produce alterations in the data.

a) First, we must establish that the signals presented in Figures 7 and 8 were not done as a change in the perievent signal. z-scores were calculated in two-minute windows around each collected sample of the Gcamp signal (one minute in front and one minute behind each point in time). Individual trials only last a few seconds. So when mice are engaged with the task, each data point is z-scored across multiple trials, not 1. This can also be observed in Figure 7C above in which the animal completes 4 trials in a 60 second time period. We performed this z-scoring for the following reasons:

  • Using this moving window method for z-scoring helped to account for changes in the raw GCamp signal over the course of an individual hour-long session. Along with the de-bleaching methods described above, this method of z-scoring helped to produce a stationary time-series for analysis.

  • By z-scoring data, we could confidently average Gcamp signals over multiple days and multiple mice.

  • Very similar photometry methods have been employed previously in the literature (see4).

b) We further note that when the reviewer says “larger peak response around the larger outcome” is established in the literature that:

  • Our data is consistent with this outcome. Our data demonstrate more dynamic initiation peaks after large rewards. We again emphasize the difference between peak magnitude and peak prominence. Initiation peaks subsequent to a large reward are more dynamic than after a small reward – they have a larger prominence than after small reward outcomes. This larger response occurs against a more silent immediate background, suggesting a greater signal to noise after large reward than after small reward. The relationship between the reward signal and the immediate background signal may be a critical component of encoding reward in this time epoch.

  • While unit recordings show increases in activity with reward, we note again that the object of our analysis is temporally constrained. We are examining initiation events that follow large and small reward outcomes. As such, we are analyzing events that lag reward consumption (this is appropriate as the majority of our behavioral analysis looks at actions that lag outcome by 1 trial). In fact, given the presence of large peaks in our data (such as those seen in Figure 7C and acknowledged by the reviewers in their previous critique), it is possible that larger peak responses may be temporally linked to epochs that occur during and after reward consumption. We might expect larger exit peaks after big rewards for instance.

  • It is unclear whether this comment: “Nearly all of the work on striatal populations has shown that these circuits scale with reward value” is referring to activity at task outcome, but if so, there are many examples where this is not the case, both for single unit recordings (see Figure 5 in5) and SPN subtype-specific photometry (see Figure 10 in4).

c) The main reason that we z-score is to compare different animals across multiple sessions. Not z-scoring, adds more noise to average traces, without changing the relative dynamics of those signals.

Author response image 5
Comparison of raw averaged signals (A) and z-scored signals (B).

The data above clearly demonstrates that the effect of z-scoring is to increase the signal to noise, rather than decrease it. This is mostly a result of allowing us to average the data from multiple mice on multiple days, but there is another effect as well. We have established that there is a moving baseline signal that varies based on the engagement of the animal with the task. Note that when we calculate ΔF/F, we do so using the entire trace. That is, we include both engaged and disengaged periods of the task. z-scoring locally, rather than globally, will thus tend to raise the relative baseline for engaged periods, not lower it, as the reviewer suggests. This is because z-scoring locally will tend to exclude the disengaged periods. The end results are signals with larger magnitude and prominence, as well as less variation, as demonstrated above.

3) Please address the below concern by tempering the language in the manuscript.

Although the authors address concerns regarding specificity of the NEX-Cre/Nrx1aKO mouse line and potential mechanistic link between forebrain function in the text, there remains a large disconnect between the widespread loss of neurexin1a expression and the specific effects seen in the DMS->SNr circuitry. As the authors mention, the NEX-Cre has primary recombination sites in the amygdala, hippocampus, and several PFC subregions; each which feed into the DMS-> SNr pathway and each provide a unique aspect to reward-associated behavior. A key premise of the manuscript is that loss of forebrain neurexin1a leads to disruption of the DMS->SNr circuitry during reward-seeking. However, what is missing from the manuscript is how is this circuit disruption occurring. Given the widespread loss of neurexin1a expression, it is difficult to infer a potential mechanism or pathway as a source for DMS activity disruption, beyond likely glutamatergic input.

We have tempered the language and agree that a disconnect remains between the brain region-specific genetic manipulations and the analysis of striatal population signals. However, this is a gap that will not be bridged in any single paper. Our goal here was to demonstrate that given the near brain-wide expression of Nrxn1a, we could in fact localize a reward-processing deficit to Nrxn1a disruption in specific brain structures – a non-trivial feat. We believe the novelty of this itself, coupled with the first analysis of a neuropsychiatric disease model via reinforcement learning paradigms and in vivo imaging of the telencephalic excitatory knockout mice are notable first steps. Furthermore, while we now cite work demonstrating Nrxn1a-associated changes in excitatory transmission within prefrontal- and thalamo-striatal circuits, it would take substantially more work – beyond the scope of this paper – to connect this type of ex vivo acute slice analysis with the altered photometry signals we observe in NexCre; Nrxn1afl/fl mice.

4) Please address the below concerns by performing an additional experiment to examine mEPSCs.

A main lingering concern is still about the retroAAV cre injection into SNr. The mIPSC recordings in DMS show that one particular change (in inhibitory transmission) does not occur but doesn't rule out other possibilities, particularly in terms of potential changes in excitatory transmission. It is assumed that the mEPSCs in Nex-cre Nrxn1a-cKO mice change in comparison to Nex-cre controls (that is their hypothesis, that corticostriatal deficits occur and underlie the changes in dSPN activity), although they don't show it but reference unpublished data in the reviewer response. A retroAAV cre injection in Nrxn1a cKO-only animals (no Nex-cre) to determine that mEPSCs don't change if the Nex-cre isn't there would be important. This would show that retroAAV cre expression doesn't do anything additional that is not intended at their synapse of interest.

We agree that the complex interconnectedness of the BG is a challenging issue given the manipulation that we had to use to isolate single SPN population signals in the context of our NexCRE;Nrxn1 mutant mice. As such, we have performed the requested experiment to discern whether adult deletion of Nrxn1a from retroAAV2-Cre injections into SNr changes the excitatory synaptic tone onto dorsal striatal dSPNs, the population imaged with photometry in Figures 7 and 8. Into adult Nrxn1lfl/fl mice, we injected rAAV2-3xFlag-Cre (plus DIO-tdTOM reporter virus in dorsal striatum) or rAAV2-EGFP-∆Cre (an enzymatically inactive truncated version of Cre with its own fluorophore) into the SNr, waited the same duration of time as for our behavior/recording analysis (3 weeks) and then recorded mEPSCs in acute slices cut from these mice. We did not observe any significant change in the frequency or amplitude of mEPSCs in neurons with Cre or δ-Cre expression, making it unlikely that adult-mediated alterations in excitation resulting from our retro-AAV2-Cre strategy account for changes in recorded Ca2+ signal. These data have been added to Figure 8—figure supplement1. Together with the prior revision’s mIPSC data, we have made a solid case that our technical approach for imaging has not dramatically contributed to the imaging results in Figure 8. We still believe the most straight-forward explanation of our data is that deletion of Nrxn1a from cortical progenitors somehow alters the excitatory synapses of these projections pathways, with one downstream result being the abrogation of value-related signals in a key target, the DMS. We have added a reference to our bioRxiv paper (currently under revision), focused on excitatory synaptic transmission in mPFC- and thalamo-striatal synapses of Nrxn1a ex vivo brain slices. However, we also clarify that given the broad cortical deletion of Nrxn1a with the NexCre, we are hesitant to try directly linking these results. In fact, as we write in the final paragraph, it will take in vivo recordings of multiple circuit nodes to fully understand how altered striatal signals come about in these mutants – work that is planned for the future.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Recommendations:

It is suggested that the data presented in the response to reviewers be accessible so that the readers can see the raw data minimally processed. A concern hinges upon data that is included in the response where the z-scored data is introducing changes into the data that is not seen in the raw df/f data. Author response image 5 clearly shows that there are significant differences between A – unprocessed data and B- z-scored data that makes it look like huge decreases are happening that do not seem apparent in the unprocessed data. While this could be fine if the authors are transparent with their analysis pipeline, providing the data would help the reader draw their own conclusions.

It is recommended that the authors include this data or make it accessible to the readers in some form (i.e. link, website, etc).

We agree that full transparency is the best solution here. Along those lines, we have posted the entire dataset for this manuscript on Dryad (Alabi, Opeyemi (2020), Neurexin Photometry, Dryad, Dataset, https://doi.org/10.5061/dryad.vhhmgqnrq). Together with the relevant Matlab code posted on the publicly accessible Fuccillo lab Github site (https://github.com/oalabi76/Nrxn_BehaviorAndAnalysis), and the more detailed methods section in the revised manuscript, it should be possible to both recreate our exact analyses, as well as perform any other desired analyses. We have added this information to the modified manuscript.

References

1 Holly, E. N. et al. Striatal Low-Threshold Spiking Interneurons Regulate Goal-Directed Learning. Neuron, doi:10.1016/j.neuron.2019.04.016 (2019).

2 Cai, L. X. et al. Distinct signals in medial and lateral VTA dopamine neurons modulate fear extinction at different times. eLife 9, doi:10.7554/eLife.54936 (2020).

3 Matias, S., Lottem, E., Dugue, G. P. & Mainen, Z. F. Activity patterns of serotonin neurons underlying cognitive flexibility. eLife 6, doi:10.7554/eLife.20552 (2017).

4 London, T. D. et al. Coordinated Ramping of Dorsal Striatal Pathways preceding Food Approach and Consumption. The Journal of neuroscience : the official journal of the Society for Neuroscience 38, 3547-3558, doi:10.1523/JNEUROSCI.2693-17.2018 (2018).

5 Shin, J. H., Kim, D. & Jung, M. W. Differential coding of reward and movement information in the dorsomedial striatal direct and indirect pathways. Nat Commun 9, 404, doi:10.1038/s41467-017-02817-1 (2018).

https://doi.org/10.7554/eLife.54838.sa2

Article and author information

Author details

  1. Opeyemi O Alabi

    1. Department of Neuroscience, Philadelphia, United States
    2. Neuroscience Graduate Group, Perelman School of Medicine, Philadelphia, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Funding acquisition, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6732-2897
  2. M Felicia Davatolhagh

    1. Department of Neuroscience, Philadelphia, United States
    2. Neuroscience Graduate Group, Perelman School of Medicine, Philadelphia, United States
    Contribution
    Data curation, Formal analysis, Investigation
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0607-5164
  3. Mara Robinson

    Department of Neuroscience, Philadelphia, United States
    Contribution
    Data curation, Investigation
    Competing interests
    No competing interests declared
  4. Michael P Fortunato

    Department of Neuroscience, Philadelphia, United States
    Contribution
    Data curation, Investigation
    Competing interests
    No competing interests declared
  5. Luigim Vargas Cifuentes

    1. Department of Neuroscience, Philadelphia, United States
    2. Neuroscience Graduate Group, Perelman School of Medicine, Philadelphia, United States
    Contribution
    Data curation, Formal analysis, Investigation
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0831-0543
  6. Joseph W Kable

    Department of Psychology, University of Pennsylvania, Philadelphia, United States
    Contribution
    Software, Formal analysis, Supervision, Writing - review and editing
    Competing interests
    No competing interests declared
  7. Marc Vincent Fuccillo

    Department of Neuroscience, Philadelphia, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    fuccillo@pennmedicine.upenn.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6569-706X

Funding

National Institutes of Health (R00MH099243)

  • Marc Vincent Fuccillo

National Institutes of Health (R01MH115030)

  • Marc Vincent Fuccillo

National Institutes of Health (F31MH114528)

  • Opeyemi O Alabi

Children's Hospital of Philadelphia (IDDRC Young Investigator)

  • Marc Vincent Fuccillo

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by grants from the NIMH (F31-MH114528 to OA, R00-MH099243 and R01-MH115030 to MVF), the Whitehall Foundation, and the IDDRC at the Children’s Hospital of Philadelphia. We thank Boris Heifets and Elizabeth Steinberg for assistance with initial Matlab code for photometry analysis. We also thank Alexandria Cowell for excellent assistance in mouse colony genotyping. Finally, we thank Patrick Rothwell and members of the Fuccillo lab for comments on the manuscript.

Ethics

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#805643) of the University of Pennsylvania.

Senior Editor

  1. Michael J Frank, Brown University, United States

Reviewing Editor

  1. Mary Kay Lobo, University of Maryland, United States

Reviewer

  1. Talia Lerner, Northwestern University, United States

Publication history

  1. Received: January 2, 2020
  2. Accepted: December 3, 2020
  3. Accepted Manuscript published: December 4, 2020 (version 1)
  4. Version of Record published: December 24, 2020 (version 2)

Copyright

© 2020, Alabi et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 501
    Page views
  • 64
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    María Fernanda López-Gutiérrez et al.
    Research Article

    Previous studies have related pair bonding in Microtus ochrogaster, the prairie vole, with plastic changes in several brain regions. However, the interactions between these socially-relevant regions have yet to be described. In this study, we used resting state magnetic resonance imaging to explore bonding behaviors and functional connectivity of brain regions previously associated with pair bonding. Thirty-two male and female prairie voles were scanned at baseline, 24h and 2 weeks after the onset of cohabitation By using network based statistics, we identified that the functional connectivity of a cortico-striatal network predicted the onset of affiliative behavior, while another predicted the amount of social interaction during a partner preference test. Furthermore, a network with significant changes in time was revealed, also showing associations with the level of partner preference. Overall, our findings revealed the association between network-level functional connectivity changes and social bonding.

    1. Developmental Biology
    2. Neuroscience
    Alessia Caramello et al.
    Research Article Updated

    During embryonic development, radial glial cells give rise to neurons, then to astrocytes following the gliogenic switch. Timely regulation of the switch, operated by several transcription factors, is fundamental for allowing coordinated interactions between neurons and glia. We deleted the gene for one such factor, SOX9, early during mouse brain development and observed a significantly compromised dentate gyrus (DG). We dissected the origin of the defect, targeting embryonic Sox9 deletion to either the DG neuronal progenitor domain or the adjacent cortical hem (CH). We identified in the latter previously uncharacterized ALDH1L1+ astrocytic progenitors, which form a fimbrial-specific glial scaffold necessary for neuronal progenitor migration toward the developing DG. Our results highlight an early crucial role of SOX9 for DG development through regulation of astroglial potential acquisition in the CH. Moreover, we illustrate how formation of a local network, amidst astrocytic and neuronal progenitors originating from adjacent domains, underlays brain morphogenesis.