Aberrant encoding of event saliency in the orbitofrontal cortex following loss of the psychiatric-associated circular RNA, circHomer1

Abstract

CircHomer1 is an activity-dependent circular RNA (circRNA) isoform produced from back-splicing of the Homer1 transcript. Homer1 isoforms are well-known regulators of homeostatic synaptic plasticity through post-synaptic density scaffold regulation. Homer1 polymorphisms have been associated with psychiatric diseases including schizophrenia (SCZ) and bipolar disorder (BD). Postmortem tissue from patients with SCZ and BD displayed reduced circHomer1 levels within the orbitofrontal cortex (OFC), a region that tracks event saliency important for modulating behavioral flexibility. While dysregulation of circHomer1 expression has recently been identified across multiple psychiatric and neurodegenerative disorders and is associated with impaired behavioral flexibility in mice, it is unknown whether circHomer1 can induce electrophysiological signatures relevant to cognitive dysfunction in these disorders. To examine the role of circHomer1 in neuronal signaling, we bilaterally knocked down circHomer1 in the OFC of C57BL/6 J male mice and recorded neural activity from the OFC during a touchscreen reversal learning task then measured molecular changes of synaptic regulators following knockdown. Knockdown of circHomer1 within the OFC induced choice-dependent changes in multiunit firing rate and local field potential coordination and power to salient stimuli during reversal learning. Further, these electrophysiological changes were associated with transcriptional downregulation of glutamatergic signaling effectors and behavioral alterations leading to impaired cognitive flexibility. CircHomer1 is a stable biomolecule, whose knockdown in rodent OFC produces lasting electrophysiological and transcriptional changes important for efficient reversal learning. This is, to our knowledge, the first demonstration of a psychiatric-associated circRNA contributing to electrophysiological, transcriptional, and behavioral alterations relevant to psychiatric phenotypes.

Introduction

Schizophrenia (SCZ) and Bipolar Disorder (BD) are heterogeneous disorders that combined affect approximately 2–3.5 percent of the population [1, 2]. While considered distinct disorders, there is a high degree of overlap in symptoms and genetic factors that predispose an individual to SCZ and BD [3–7]. Both disorders are highly polygenic with an enrichment of genes involved in synaptic plasticity implicating synaptic dysfunction as an underlying mechanism [3–10].

Homeostatic synaptic plasticity is an intricate mechanism whereby synapses balance their respective weights to evenly tune a circuit and prevent hyperexcitability within the complex system [11–14]. HOMER1 is a homeostatically regulated synaptic scaffold protein [15] that has been previously linked to psychiatric disease and synaptic signaling [9, 16–20]. Polymorphisms in this gene have been associated with psychiatric phenotypes in genome-wide association studies [6, 20, 21]. The Homer1 gene undergoes alternative splicing to encode for multiple postsynaptic density (PSD) proteins [22, 23] that interact through a scaffolding structure with surface N-methyl-D-aspartate receptor (NMDA) receptors [24], type 1 metabotropic glutamate receptors (mGluR1/5) [25–27], and intracellular proteins including SHANK [24] and inositol 1,4,5-trisphosphate receptor (IP₃Rs) to control intracellular Ca²⁺ signaling [28] (see review [18]). Long isoforms, HOMER1B/C, are constitutively expressed and contain an EVH1 domain to interact with other PSD proteins and a coiled-coil domain that allows for the formation of tetrameric complexes important for protein scaffolding [18, 24]. The short, activity-dependent isoform, Homer1a, is transcribed during heightened network excitability [23, 29] and acts as a dominant negative to displace the long isoforms by interacting through its EVH1 domain [29]. However, HOMER1A lacks the coiled-coil domain and cannot dimerize, effectively breaking down PSD scaffolds and reducing postsynaptic signaling [29] through a homeostatic downscaling mechanism involving α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) internalization [15, 30]. Recently, circHomer1, another activity dependent Homer1 splice isoform was identified [31–33]. CircHomer1 is a circular isoform created through back-splicing in which exons 2 and 5 covalently join to form a closed-loop structure [34, 35] that has not been shown to be translated into protein, but the RNA is able to bind other RNA species and proteins [31, 32, 36, 37] suggesting a more complex regulatory role. CircHomer1 is an inducible isoform [33] with RNA expression similar to that of Homer1a [31]. Reduction of circHomer1 relative to controls has been found in numerous psychiatric and neurological disorders including SCZ and BD [31, 32], Alzheimer’s Disease [38–40] and frontotemporal lobe degeneration [40], mesial temporal lobe epilepsy [41], and Parkinson’s Disease [42]. Postmortem brain samples from the orbitofrontal cortex (OFC) of patients with SCZ and BD showed a significant reduction of circHomer1 expression relative to healthy controls that was correlated with disease onset and psychosis status [31, 32] suggesting expression may represent disease subtypes and severity.

The OFC is a frontocortical region associated with outcome-monitoring involved in goal-directed action selection during decision-making [43–45]. Through its reciprocal connectivity with the striatum, the OFC tracks when reward contingencies change and encodes the value of the reward [45–48] in its particular circumstance prompting either a value-guided or habitual response through its interactions with dorsal medial or dorsal lateral striatum, respectively [44, 48, 49]. Dysfunction of this region is commonly found in individuals with psychiatric phenotypes including BD and SCZ and is believed to underlie some of the detrimental behaviors associated with poor decision-making in these populations [50, 51].

Recent work has uncovered electrophysiological signatures that correlate with neural dysfunction in patients with SCZ and BD [52, 53]. Importantly, similar signatures have been found in animal modeling of traits seen in these disorders including psychosis-like states and cognitive dysfunction [54, 55]. Cognitive-associated electrophysiological signatures are conserved across species [53, 54] and provide a measurable representation of network changes underlying cognitive dysfunction. Our previous work [32] demonstrated knockdown of circHomer1 in the OFC specifically disrupts mid-stage (chance) reversal learning, the decision phase of behavioral flexibility, and predicted a disruption in synaptic activity using RNA sequencing and genetic pathway analysis; yet, to our knowledge, no studies have identified a psychiatric disease-relevant circRNA capable of regulating neuronal network activity and coordination during behavior. To determine the role of circHomer1 in neuronal signaling within the OFC during learning, we paired targeted knockdown of circHomer1 within mouse OFC with in vivo electrophysiological recordings during a touchscreen visual reversal task and measured coordinated neural activity and expression of synaptic genes.

Materials and methods

For more detailed methodology, see Supplemental Experimental Procedures.

Ethics approval

All experimental procedures were performed in accordance with the guidelines and regulations of the Institutional Care and Use Committee (IACUC) at the University of New Mexico Health Sciences Center under protocol 24-201564-HSC.

Animals

WT mice used in our study were adult male C57BL/6 J mice (The Jackson Laboratory, Bar Harbor, ME) aged 2–5 months. Investigators were blinded where possible to animal treatment groups.

Operant behavioral testing and stereotaxic surgery

Twenty male C57BL/6 J mice (n = 10 per group) were initially acquired from The Jackson Laboratories and used for in vivo electrophysiological testing. Numbers of animals used for behavior, electrophysiology, and molecular characterization are described throughout and in more detail in Supplementary Experimental Procedures. Behavioral acclimation and operant testing were performed as described previously [31, 32]. Following pre-training, animals underwent stereotaxic surgery, which involved delivery of either a non-specific scrambled sh-control lentiviral vector or an sh-circHomer1 lentiviral vector targeted to the unique backsplice junction of exons 2 and 5 of the mature circHomer1 RNA product [31, 32].

In vivo electrophysiological recordings

Neuronal activity was recorded using a custom 16-channel (2x2x4) fixed microarray with OFC-specific separation (2.75 mm) (Innovative Neurophysiology, Inc.) and collected using a digital multichannel acquisition processor (OmniPlexD, Plexon, Dallas, TX, USA), as previously described [48, 55, 56] during seven sessions throughout discrimination and reversal on the following days: day two of discrimination, discrimination criterion (85% correct), first day of reversal (R1), three consecutive days at chance reversal (CR1-3 - starting on the day after the animal performed at least 10/30 correct trials), and reversal criterion (R85 - the day following the animal achieving 85% correct).

Tissue harvesting and RNA extraction/circRNA quantification

Following completion of behavioral testing, OFC micropunches were obtained and processed as described previously [31, 32] for mRNA and circRNA quantification by qRT-PCR.

Results

KD of circHomer1 in the OFC of mice during a visual discrimination and reversal paradigm disrupts behavior during mid-stage reversal

Harnessing a translational touchscreen approach paired with in vivo electrophysiological recordings, we captured real-time activity from excitatory neurons within the OFC while an animal was performing an executive function task. Using lentiviral-mediated expression (Fig. 1A, B), we employed shRNA knockdown of circHomer1 within the bilateral OFC (Fig. 1A), which was previously shown to impair mid-stage (chance) reversal learning and alter gene isoform expression favoring a reduction in synaptic activity [32]. Using an asymmetrically designed shRNA targeting the unique circHomer1 backsplice junction [32] of exons two and five (Fig. 1B), we obtained an approximately 40 percent reduction in the mature circHomer1 transcript within the OFC (Fig. 1E). The shRNA is driven by a ubiquitous mouse U6 promoter and delivered locally to the OFC (Supplementary Fig. 1A), which predominantly consists of cortico-cortical and corticostriatal glutamatergic pyramidal neurons [57, 58]. Indeed, we observe viral expression in projection neurons in sensory-motor cortices as well as the dorsal striatum (Supplementary Fig. 1B). Viral delivery and electrode implantation occurred after initial training to ensure all animals appropriately learned to perform the touchscreen task (Fig. 1D). Bilateral microarrays were implanted into the OFC (AP + 2.6, ML, +/− 1.35, DV −2.6) with eight separate wires per hemisphere spaced 150 µm apart to increase the likelihood of capturing distinct neuronal units [59] (Fig. 1A). Following two weeks of recovery allowing for stable viral expression and acclimation to the electrode implant, mice were trained to the initial discrimination, in which they had to correctly choose one shape (Fig. 1C, “Discrimination”) 85 percent of the time. Once criterion was achieved, they were switched to reversal and had to re-attain 85 percent correct responses on the previously unrewarded shape (Fig. 1C, “Reversal”). Animals were recorded seven times across the discrimination and reversal paradigm (Fig. 1D).

Fig. 1. sh-RNA-mediated knockdown of circHomer1 and in vivo microarray recording in the OFC during touchscreen visual reversal task.

A Schematic (left) of 16-channel microarray implanted into bilateral OFC. Representative histological section of OFC following electrolytic lesion to mark electrode placement. GFP marks lentiviral expression. B Lentiviral vector carrying sh-RNA for circHomer1. C Touchscreen paradigm from Reward Habituation through Discrimination-Reversal. D Experimental timeline. E. Mean ± SEM expression of circHomer1 within the OFC relative to scramble-injected control animals. All data were normalized to 18 s rRNA. *p < 0.05, two-tailed one sample t-test relative to normalized control mean (t₍₃₎ = 5.69, p = 0.01). (n = 6 control, 4 circHomer1 KD).

Correct trials were rewarded with strawberry Nesquik (Supplementary Fig. 2A), and incorrect trials were punished with a ten-second timeout and illumination of a houselight (Supplementary Fig. 2A). In recorded animals, no significant difference was found in the number of sessions to obtain the initial discrimination between circHomer1 KD animals and controls (Supplementary Fig. 2B). However, in those animals with confirmed KD of circHomer1, significantly (p < 0.05) more correct (Fig. 2A), incorrect (Fig. 2B), and correction (Fig. 2C) trials were required to move through chance reversal (33-66% correct) compared to controls, while early and late reversal stages were not significantly impacted. Additional analysis of specific trial types (Supplementary Fig. 2C, D) during chance reversal confirmed previous work [32] indicating reduction of circHomer1 in the OFC significantly (p < 0.0001) impairs chance reversal measured by an increase in the number of all trial types during chance reversal (Supplementary Fig. 2D). We previously showed circHomer1 expression negatively correlates with behavioral performance during chance reversal (i.e. lower circHomer1 led to an increase in the number of trials). Our current work found a similar result where circHomer1 expression negatively correlated with behavioral performance during mid-to-late reversal (Fig. 2D). Looking across all trials and behavioral stages (early, chance, and late reversal), we found lower circHomer1 expression is associated with fewer trials during early reversal but more trials during mid-to-late reversal, specifically more incorrect/correction trials (Fig. 2D). Together, these data demonstrate circHomer1 expression within the OFC is predictive of reinforcement learning behavior.

Fig. 2. Behavioral performance during reversal learning in animals with confirmed circHomer1 KD.

A Total correct trials summed across sessions shown as mean ± SEM for early, chance, and late reversal. B Total incorrect trials summed across sessions shown as mean ± SEM for early, chance, and late reversal. C Total Correction trials summed across sessions shown as mean ± SEM for early, chance, and late reversal. D Spearman correlation between reversal trial types (correct, incorrect, correction) and circHomer1 mRNA expression in combined set of KD and control animals that had complete data for each session and mRNA expression (n = 7). In (A–C), Asterisks indicate adjusted P-value following two-way ANOVA with Bonferroni’s correction for multiple comparisons. For (A–C) (n = 4 circHomer1 KD and 6 controls). In (D), color and size indicates direction and Spearman correlation. P-value: *< 0.05, **< 0.01, ***< 0.001.

Knockdown of circHomer1 within the OFC dynamically regulates choice-responsive firing rates

To determine neuronal responsiveness to stimuli that may contribute to this learning impairment, we analyzed neuronal firing rates across a four second (1 second pre- to 3 second post-choice) epoch for correct and incorrect choices (Fig. 3A, B). For correct choices, a robust increase in event-responsive firing was observed in the one-second post-correct choice epoch, in which a secondary reinforcing tone is played to signal a correct choice (Fig. 3C–G). Firing rates from both circHomer1 KD animals and controls increased relative to the one second pre-correct choice baseline during the conditioned stimulus (CS) tone period across all reversal recording days (Fig. 3C–G). Alternatively, for incorrect choices, neurons from both groups showed a slow ramping in activity following the incorrect choice (Fig. 3H, I). On the first day of reversal (R1), the average OFC neuronal firing rate from circHomer1 KD animals was increased significantly (p < 0.0001) over controls in the one-second post-correct choice window when the CS is presented before significantly dropping below control animals in the two- (p < 0.05) and three- (p < 0.0001) second post-choice epochs, when the reward is consumed. (Fig. 3C). Similarly, for mid-chance reversal (CR1-3) and criterion reversal (R85) sessions, firing rates remained elevated in the 2–3 second post-correct choice epoch for OFC neurons from control animals compared to those from circHomer1 KD animals (p < 0.0001) (Fig. 3D–G). Only on the second day of chance reversal (CR2) did correct choice-responsive firing rate from OFC neurons in control animals significantly (p < 0.001) increase over circHomer1 KD during the CS period (Fig. 3E). These findings demonstrate a shift in correct-choice responsive neural firing from reward-responding to cue-responding in circHomer1 KD animals.

Fig. 3. Reward- and punish-associated in vivo neuronal firing across reversal learning.

Analysis windows for reward- (A) and punish- (B) associated neuronal firing during touchscreen visual reversal learning. Correct choices are reinforced by a 1-second tone, while incorrect choices are punished with a 10-second timeout and houselight presentation. Mean ± SEM of OFC firing rates as 50 ms bins for correct choices on the first day of reversal (n = 218 control, 306 circHomer1 KD units) (C), first day of chance reversal (n = 210 control, 332 circHomer1 KD units) (D), second day of chance reversal (n = 217 control, 327 circHomer1 KD units) (E), third day of chance reversal (n = 248 control, 320 circHomer1 KD units) (F) and reversal criterion (n = 288 control, 266 circHomer1 KD units) (G). Mean ± SEM of OFC firing rates as 50 ms bins for incorrect choices on the first day of reversal (n = 223 control, 305 circHomer1 KD units) (H), first day of chance reversal (n = 211 control, 326 circHomer1 KD units) (I), second day of chance reversal (n = 198 control, 324 circHomer1 KD units) (J), third day of chance reversal (n = 247 control, 317 circHomer1 KD units) (K) and reversal criterion (n = 288 control, 266 circHomer1 KD units) (L). M Mean ± SEM of one-second pre-choice baseline firing rate across reversal sessions. N Mean ± SEM of anticipatory (100 ms) pre-choice firing rate across reversal sessions. O Percent event-responsive units for correct choice across reversal sessions. P Percent event-responsive units for incorrect choice across reversal sessions. Reversal day 1 (R1), chance reversal day 1 (CR1), chance reversal day 2 (CR2), chance reversal day 3 (CR3), reversal criterion (R85). P-value: *‡< 0.05, **‡‡< 0.01, ***‡‡‡< 0.001, ****‡‡‡‡< 0.0001 represents separate two-way ANOVAs across 1-sec averaged bins for correct and incorrect choices with Bonferroni’s post-hoc multiple comparisons test. For (M) and (N): black = control correct trials (TC), gray = control incorrect trials (TiC), dark red = circHomer1 TC, light red = circHomer1 (TiC).*Represents significance for correct choices. ‡Represents significance for incorrect choices. Chi-square contingency test was used in (O) and (P) for each reversal session.

For incorrect choices, average firing rate of OFC neurons from circHomer1 KD animals significantly (p < 0.05) increased over controls on the first day of reversal (Fig. 3H) for the entire three-second post-choice epoch. However, firing rate of OFC neurons from circHomer1 KD animals significantly dropped below controls on the first (p < 0.001) (Fig. 3I) and second day of chance reversal (Fig. 3J) during the post-incorrect choice epoch before increasing again over controls on the last day of reversal (p < 0.05) (Fig. 3L). A modest reduction in firing rate (p < 0.05 for the end of the post-choice window) was seen on chance reversal day 2 (Fig. 3J), while no differences were seen on chance reversal day 3 (Fig. 3K). Together, these results confirm previous work demonstrating firing rates of OFC neurons are dynamic across learning stages [48, 55] and provide novel evidence that they are differentially modulated following a choice when circHomer1 expression is reduced.

Because baseline firing rate can be indicative of synaptic properties and intrinsic excitability [60], we compared baseline firing frequency from the one-second pre-choice epoch across reversal sessions and found that it, too, was dynamically regulated throughout reversal (Fig. 3M). Baseline firing rates differed significantly between groups on chance reversal days 1 (p < 0.0001, correct choice; P < 0.001, incorrect choice) and 2 (p < 0.01 correct choice; p < 0.05 incorrect choices) and criterion reversal (p < 0.001 correct choice; p < 0.0001 incorrect choices) for the one-second epoch preceding choices (Fig. 3M) demonstrating differential rate modulation of neuronal firing frequency across learning stages that is impacted by reduction of circHomer1 within OFC neurons. Firing rate modulation in the immediate window preceding a choice is associated with attention allocated to the choice and anticipation of a predicted outcome [45]. For this reason, we analyzed firing rate in the 100 ms immediately before a choice when the animal would be oriented to the screen and found similar rate modulation to the entire baseline window across reversal. However, the firing rate within this 100 ms window was higher than the baseline firing rate across the entire one second baseline window (Fig. 3N). Anticipatory firing rate was significantly higher in OFC neurons from circHomer1 KD animals on the first and last day of reversal for correct choices (p < 0.001) and incorrect choices (p < 0.0001) (Fig. 3N) when the outcome-associations are less ambiguous. These results suggest circHomer1 is important for maintaining baseline firing characteristics of OFC neurons.

While the firing rate of neuronal populations is likely to change in response to a salient event, the proportion of OFC neurons recruited to respond to a specific event has also been shown to be dynamically regulated throughout learning [48]. To determine whether neuronal units were differentially recruited in KD and control animals across reversal stages, we analyzed the percentage of neuronal units as a fraction of the total that significantly (p < 0.05) increased or decreased their firing rate with respect to the pre-choice baseline across reversal learning stages. Significant differences between groups arose on the second day of chance reversal, whereby control animals had a significantly (p < 0.0001) higher proportion of correct choice-responsive units compared to circHomer1 KD animals, and this difference was maintained throughout the remaining reversal learning sessions (Fig. 3O). Oppositely, the percentage of incorrect choice-responsive units was significantly higher in circHomer1 KD animals compared to controls on chance reversal days 2 (p < 0.05) and 3 (p < 0.0001), while the proportion of incorrect-responsive units was significantly reduced in circHomer1 KD animals compared to controls on the final session of reversal learning (p < 0.0001) (Fig. 3P). These results indicate that reduction of circHomer1 within the OFC alters neuronal recruitment to a choice through enhanced error-responsive recruitment and reduction of correct-responsive recruitment during chance reversal learning.

To further confirm a direct role for circHomer1 in pyramidal neuron firing properties we knocked down circHomer1 in cultured mouse hippocampal pyramidal neurons, a widely used and validated cellular model in which circHomer1 has been previously examined [33]. No significant changes were found in inward sodium (Supplementary Fig. 3A, B) or outward potassium current (Supplementary Fig. 3A, C). We measured spontaneous excitatory postsynaptic currents (sEPSCs) at a holding potential of -70mV to capture AMPAR-mediated responses and observed a significant reduction in sEPSC frequency (p < 0.05) (Supplementary Fig. 3D, E) with no change in amplitude (Supplementary Fig. 3F). This result supports in vivo data showing circHomer1 KD blunts baseline neuronal firing rate and suggests circHomer1 is important for spontaneous, AMPAR-mediated neuronal firing. Altogether, these data implicate a role for circHomer1 in maintenance of both baseline neuronal characteristics at the cellular and network level as well as in an evoked response to salient stimuli during reinforcement learning.

ITPC is aberrantly increased across reversal learning following circHomer1 KD in the OFC

Local field potential (LFP) oscillatory phase alignment was assessed across trials to determine event-locked shifts in coordinated oscillatory activity in response to correct and incorrect choices. Inter-trial phase consistency (ITPC) is a measure of the coordinated activity of a neuronal population and is thought to be important in the processing of salient events [61–63]. An ITPC value of one represents the phase of oscillations aligning perfectly at a given time, while a value of 0 represents no phase alignment. ITPC was aligned to correct and incorrect choices and analyzed within the same one second pre-choice and three second post-choice window as used for spike-firing analysis. In accordance with previous work [48, 55], frequency-specific region of interest (ROI) analyses were performed across the following frequency ranges: low delta (1–2 Hz), theta (4–8 Hz), alpha (9–13 Hz), beta (14–30 Hz), and gamma (30–50 Hz) during the 0–250 ms post-choice epoch (Fig. 4B, black boxes) to examine phase alignment during a choice-responsive time window in which ITPC is normally modulated in wild-type animals. As expected, correct-choice associated ITPC peaked early in chance reversal (Fig. 4A–E, Supplementary Fig. 4A–E) and dissipated for the remaining reversal sessions for all tf-ROIs.

Fig. 4. Choice-responsive intertrial phase consistency (ITPC) across reversal stages.

Correct choice ITPC for control (top) and circHomer1 KD animals (bottom) on reversal day 1 (A), chance reversal day 1 (B), chance reversal day 2 (C), chance reversal day 3 (D) and reversal criterion (E). Time-frequency ROI (tf-ROI) analysis at each frequency band for correct choices on the first day of reversal shown as mean ± SEM (n = 6 control, 9 circHomer1 KD) (F), averaged across the three chance reversal sessions (n = 8 control, 9 circHomer1 KD) (G), and reversal criterion (n = 6 control, 7 circHomer1 KD) (H). Incorrect choice ITPC for control (top) and circHomer1 KD animals (bottom) on reversal day 1 (I), chance reversal day 1 (J), chance reversal day 2 (K), chance reversal day 3 (L) and reversal criterion (M). tf-ROI analysis at each frequency band for incorrect choices on the first day of reversal shown as mean ± SEM (n = 6 control, 9 circHomer1 KD) (N), averaged across the three chance reversal sessions (n = 8 control, 9 circHomer1 KD) (O), and reversal criterion (n = 6 control, 7 circHomer1 KD) (P). Black rectangles in (B) outline choice-associated (0-250 ms) ROIs for (F, G, H, N, O, P). P-value: #< 0.1, * < 0.05, **< 0.01, ***< 0.001, ****< 0.0001 represents two-way ANOVA (brackets indicate main effect of group). Significant post-hoc (Bonferroni) multiple comparisons tests are indicated by asterisks over appropriate frequency band.

No group differences were observed for correct choice-responsive ITPC on the first (Fig. 4A, F) and final (Fig. 4E, H) session of reversal across tf-ROIs. A significant increase in ITPC was observed across tf-ROIs in the OFC of circHomer1 KD animals coincident with the behavioral deficit across chance reversal sessions (Fig. 4B–D, G).

Incorrect choice-associated ITPC similarly peaked on the first day of chance reversal (Fig. 4I, Supplementary Fig. 4F–J). A modest (p < 0.1) increase in incorrect choice-associated ITPC was observed in circHomer1 KD animals on the first day of reversal (Fig. 4I, N). A significant (p < 0.0001) increase in incorrect choice-associated ITPC was found during chance reversal sessions in circHomer1 KD animals, with particularly strong effects in low frequency delta and theta tf-ROIs (Fig. 4J–L, O). By the final reversal session, ITPC associated with incorrect choices was no longer significantly increased above control levels (Fig. 4M, P). These data point to aberrantly sustained low frequency ITPC following both correct and incorrect choices across critical learning stages. The dynamic event-locked modulation of OFC ITPC across reversal stages recapitulates previous findings [55] and suggests circHomer1 is important for the proper timed coordination of oscillatory activity during learning.

Reward-related alterations in time-frequency power are present in the OFC of circHomer1 KD animals

Time-frequency power is a measure of the amplitude of a signal and has been used as an electrophysiological marker to assess cognitive control [53, 64, 65]. The same frequency range definitions from our ITPC analysis were used to assess power (Fig. 5A–E, I–M); however, two separate time-windows were analyzed for correct and incorrect choices as power fluctuations were found to differ in response to reward- (250–550 ms) (Fig. 5B, black boxes) and punishment-related (500–800 ms) (Fig. 5J, black boxes) signaling in wild-type animals [66]. No significant group differences were observed across time-frequency ROIs (tf-ROIs) on the first day of reversal (Fig. 5A, F) following correct choices. Interestingly, reward-related power within the tf-ROIs was significantly (p < 0.05) diminished during chance reversal (Fig. 5B–D, G) in circHomer1 KD animals and remained lower (p < 0.05) than controls at reversal criterion (Fig. 5E, H). Notably, data are presented as a change from the one-second pre-choice baseline (dB), with control animals exhibiting a general increase in reward-related power relative to baseline during chance and criterion reversal, while circHomer1 KD animals exhibited either decreased or unchanged reward-related power during these reversal stages (Fig. 5G-H, Supplementary Fig. 5A–E).

Fig. 5. Time-frequency choice-responsive power across reversal stages.

Correct choice-associated power for control (top) and circHomer1 KD animals (bottom) on reversal day 1 (A), chance reversal day 1 (B), chance reversal day 2 (C), chance reversal day 3 (D) and reversal criterion (E). Time-frequency ROI (tf-ROI) analysis at each frequency band for correct choices on the first day of reversal shown as mean ± SEM (n = 6 control, 9 circHomer1 KD) (F), averaged across the three chance reversal sessions (n = 8 control, 9 circHomer1 KD) (G), and reversal criterion n = 7 control, 7 circHomer1 KD) (H). Incorrect choice-associated power for control (top) and circHomer1 KD animals (bottom) on reversal day 1 (I), chance reversal day 1 (J), chance reversal day 2 (K), chance reversal day 3 (L) and reversal criterion (M). tf-ROI analysis at each frequency band for incorrect choices on the first day of reversal shown as mean ± SEM (n = 6 control, 9 circHomer1 KD) (N), averaged across the three chance reversal sessions (n = 8 control, 9 circHomer1 KD) (O), and reversal criterion (n = 7 control, 7 circHomer1 KD) (P). Black rectangles in (B) outline reward-associated (250-550 ms) ROIs for (F, G, H). Black rectangles in (J) outline punishment-associated (500–800 ms) ROIs for (N, O, P). P-value: #< 0.1, * < 0.05, **< 0.01, ***< 0.001, ****< 0.0001 represents two-way ANOVA (brackets indicate main effect of group). Significant post-hoc (Bonferroni) multiple comparisons tests are indicated by asterisks over appropriate frequency band.

Reduction of tf-ROI power relative to baseline was observed following incorrect choices consistent with previous work measuring punishment-associated power changes in rodents and humans [66] (Fig. 5I–P). Again, no significant group differences were present on the first day of reversal across the tf-ROIs measured (Fig. 5I, N). Strikingly, however, a robust difference was observed between circHomer1 KD animals and controls during chance reversal, particularly for mid-range frequencies (theta-beta) (Fig. 5J–L, O). During chance reversal, incorrect choice-responsive tf-ROI power was significantly reduced relative to baseline in controls when compared to circHomer1 KD animals (Fig. 5O). No group differences in incorrect choice-responsive tf-ROI power were measured at reversal criterion (Fig. 5M, P). In contrast to correct choice-responsive power, control animals tended to show a reduction in incorrect-choice associated power, while circHomer1 KD animals showed a markedly diminished change from their baseline power (Supplementary Fig. 5F–J). These tf-ROI power analyses are consistent with previously reported event-responsive power changes [66]. Importantly, they indicate reduction of circHomer1 within the OFC induces reward- and error-related neural alterations during chance reversal, the stage where behavior is significantly impaired in circHomer1 KD animals.

KD of circHomer1 alters the transcriptional landscape of post-synaptic signaling genes

Both our in vitro and in vivo recording data implicated a reduction in coordinated neuronal firing potential. Our previous work also demonstrated that KD of circHomer1 within the OFC alters specific isoforms important for synaptic signaling including Grin2b and Fmr1 [32]. We quantified mRNA expression for known regulators of excitatory glutamatergic neurotransmission (Fig. 6A) including immediate early gene Homer1a and PSD scaffold component Homer1b; genes encoding NMDAR subunits, Grin2a and b; genes coding for AMPAR subunits, Gria1 and 2, and Fmr1, which encodes the RNA binding protein FMRP, to identify a potential mechanism by which KD of circHomer1 is able to influence synaptic signaling. A main effect of group was observed, whereby KD of circHomer1 resulted in a general reduction of all synaptic-associated transcripts tested (p = 0.001); however, no individual transcripts were found to be significantly reduced following Bonferroni’s post-hoc correction (Fig. 6B). Because these transcripts are generally regulated in unison due to their role in excitatory neurotransmission, we expected the correlation among transcripts to be robust. As expected, we found that when all samples were combined (Fig. 6C), circHomer1 was strongly correlated with Homer1a (r = 0.90, p = 0.0004) and Homer1a and b were correlated (r = 0.71, p = 0.02). Additionally, post-synaptic targets were primarily positively correlated with each other. Fmr1 was positively correlated with Gria1 (r = 0.7, p = 0.024), Gria2 (r = 0.92, p < 0.0001), Grin2b (r = 0.87, p = 0.001), and Grin2a (r = 0.71, p = 0.02). Gria1 and Gria2 were positively correlated, (r = 0.85, p = 0.002) as were Grin2b and Gria1 (r = 0.73, p = 0.02).

Fig. 6. Gene expression changes following KD of circHomer1 in the OFC.

A Schematic of a synapse representing the interactions between HOMER1 isoforms and other synaptic modulators including NMDARs, AMPARs, and FMRP for reference. NMDARs and AMPARs consist of subunits coded by Grin2a/b and Gria1/2, respectively, depending on activity demands. Fmr1 encodes FMRP, which negatively regulates mGluR-mediated protein synthesis important for long term depression (LTD). Homer1b encodes a scaffolding isoform linking synaptic receptor components to intracellular signaling, while Homer1a encodes a dominant negative isoform, which disrupts the scaffold integrity. CircHomer1 inhibits Homer1b translocation to the synapse, which likely indirectly prevents Homer1a trafficking to the synapse. Therefore, circHomer1 likely impacts PSD receptor composition via scaffold rearrangement. B Mean ± SEM of normalized mRNA expression in total tissue from OFC micropunches. Each gene is normalized to 18S and expressed as a value relative to the control. C–E Correlation matrix of mRNA expression for all samples combined (C, n = 10) control only (D, n = 6), and circHomer1 KD (E, n = 4). Two-way ANOVA with Bonferroni’s multiple comparisons correction was used for (B). P-value represents main effect of group **< 0.01. For (C–E), Pearson’s correlation was calculated. P-value: * < 0.05, **< 0.01, ***< 0.001, ****< 0.0001. Schematic created in BioRender.

We then assessed correlation between these target genes separately for control (Fig. 6D) and circHomer1 KD animals (Fig. 6E). While the sample size is limiting, the relationships among transcripts are altered between these groups. Specifically, circHomer1 remained highly correlated with Homer1a in controls (r = 0.93, p = 0.007), but this correlation was completely lost in OFC tissue from KD animals (r = −0.009, p = 0.99), while the correlation between circHomer1 and Homer1b became negative (r = −0.35, p = 0.65). Interestingly, the direction of correlation was changed between several of the post-synaptic receptor genes in OFC tissue from circHomer1 KD animals. Grin2b became more strongly negatively correlated with Homer1a (r = −0.62, p = 0.38) and Homer1b (r = −0.85, p = 0.15) in KD animals. Moreover, Grin2a was positively correlated with Homer1 isoforms in KD animals (r = 0.74, p = 0.26 for circHomer1; r = 0.39, p = 0.61 for Homer1a; and r = 0.18, p = 0.82 for Homer1b), but negatively correlated in control animals (r = −0.71, p = 0.11 for circHomer1; r = −0.69, p = 0.13 for Homer1a; and r = −0.24, p = 0.65 for Homer1b). These correlations were not statistically significant, likely due to the small numbers of animals in each group but may indicate a compensatory change in glutamatergic receptor composition and coordination with PSD scaffolds following KD of circHomer1.

Finally, to determine whether the down regulation of these other synaptic modulators may be driving the behavioral phenotypes, we correlated all measured synaptic genes with the number of trials required during early, chance, and late reversal. For controls and KD animals combined, we found an intriguing pattern where low expression of Fmr1 and the genes encoding NMDAR/AMPAR subunits was generally associated with fewer trials during early reversal but more trials during chance-to-late reversal (Supplementary Fig. 6). Meanwhile, lower Homer1a and b expression associates with more incorrect trials early (Supplementary Fig. 6). Importantly, these data demonstrate that early reversal performance predicts mid-to-late reversal performance, where more trials during early reversal generally result in fewer trials during late reversal, and synaptic genes in the OFC differentially regulate the switch in behavioral response.

Discussion

Low expression of circHomer1 in rodent OFC impairs salience assignment important for cognitive flexibility

The goal of the current study was to determine the electrophysiological correlates resulting from loss of circHomer1 expression within the orbitofrontal cortex. Low circHomer1 expression has consistently been reported in psychiatric [31, 32] and neurodegenerative disorders [38–40, 42], and it has been reported that individuals with psychiatric disorders have behavioral manifestations of inappropriate salience signaling [67–69], whereby they assign incentive salience to cues rather than outcomes. Mistimed neuronal firing can influence inappropriate salience assignment and motivation that may drive behavioral inflexibility [55, 70] and misguide reward-seeking behavior. Altered salience signaling is a key behavioral manifestation associated with substance abuse and poor decision-making [67, 69]. Here, we demonstrate that knockdown of circHomer1 within the OFC, a frontocortical region associated with outcome-monitoring, shifts the neuronal signaling to cue-responding (i.e., increased firing rate to the tone) rather than reward-responding (increased firing rate to the reward) as observed in control animals. This heightened response to the cue appears early in reversal learning, while a loss of reward-responding occurs during chance reversal, when it is important for the OFC to monitor rewarded outcomes to update behavior [48]. The reduced firing rate of circHomer1 KD animals during the reward window coincides with behavioral impairment in these animals [32], marked by an increase in the number of trials to complete chance reversal [32], suggesting neuronal response to reward in this timeframe may be key to update value-guided behavior. We also found the proportion of correct choice-responsive neurons decreases throughout chance reversal, while the proportion of incorrect choice-responsive neurons increases in circHomer1 KD animals, indicating additional disruption in OFC neural recruitment tied to inappropriate salience assignment during a critical learning window.

Timing of population activity is thought to underlie signal propagation to downstream regions [63, 71]. The primary projections from the lateral OFC are into the dorsal striatum, a region important for producing goal-directed movement in response to outcomes [72]. Aberrant or mistimed signaling between the OFC and striatum has been shown to disrupt behavioral flexibility [55]. In the current study, we found that circHomer1 KD in the OFC resulted in aberrantly increased and sustained coordinated activity as measured by ITPC. In particular, theta ITPC, which has been shown to be especially important for long-range signal propagation [73–76] was increased in circHomer1 KD animals throughout late-stage reversal to both correct and incorrect choices. High ITPC within the OFC is associated with salient, changing event contingencies and typically peaks during early-to-mid chance reversal when the new association is being formed [48]. ITPC should, however, decrease as the outcome association is solidified indicating the outcome is no longer unexpected. Prolonged heightened ITPC is a neural marker of mismatch between outcome and expectation and the persistent increase in ITPC may prolong the chance reversal stage because the mice are not encoding the new response as correct. In addition, we observed reduced deviations in event-related tf-Power in KD animals relative to baseline, suggesting the timing of neuronal coordination is not appropriately locked to the stimulus, which may also impair behavioral updating. Given that pyramidal neurons of the OFC project onto dopaminergic neurons within the dorsal striatum, loss of circHomer1 in the OFC may alter downstream dopaminergic signaling linked to incentive salience and value-guided behavior [51, 67, 77]. Assessment of striatal activity following loss of circHomer1 will be necessary to reveal this mechanism.

Loss of circHomer1 alters the transcriptional regulation of synaptic mediators

KD of circHomer1 was previously predicted to result in reduced synaptic activity through gene expression analysis [32]; however, the molecular mechanism by which this occurs is unclear. Homer1 is homeostatically regulated through alternative splicing [23], and knockdown of the activity-dependent isoform, circHomer1, has been shown to increase both constitutive, Homer1b, and activity-dependent, Homer1a, mRNA localization into the synapse [31, 32] suggesting an influence on local translation. An increase in synaptic Homer1b has been associated with neuronal hyperexcitability in disease [78], while increased synaptic Homer1a is often associated with a homeostatic renormalization [15, 23, 30] and reduced neuronal excitability [79, 80]. Over-expression of Homer1a has also been shown to attenuate intracellular calcium release through group 1 mGluR activation [28]. Here, we measured gene expression of multiple post-synaptic genes in OFC following completion of the behavioral paradigm and found that KD of circHomer1 resulted in a global down-regulation of glutamatergic receptor gene expression (Gria1 and 2 and Grin2a and b) as well as Fmr1, which encodes the RNA binding protein FMRP responsible for regulating synaptic protein translation important for mGluR-mediated LTD [81–83]. Our results confirm the previous finding that KD of circHomer1 in the OFC results in transcriptional downregulation of key synaptic mediators [32]. However, we did not measure gene expression specifically in synaptosomes in this study due to tissue limitations following electrode removal, and synaptic transcript localization may differ from total tissue expression as we have observed previously [31]. The altered correlations of glutamatergic receptors with Homer1 isoforms in KD animals might additionally suggest a compensatory change in receptor composition to homeostatically regulate synaptic weights in an attempt to prevent prolonged hyperexcitability. It is probable that the activity-dependent transcriptional regulation of these isoforms may result in differential signaling depending on the timing and network activity demands.

Our group previously showed endogenous circHomer1 and Homer1b expression is temporally regulated across reversal learning [31]. Specifically, circHomer1 expression transiently drops in wild-type mice during chance reversal and Homer1b expression is increased at this stage, while both remain unchanged from a “baseline” state during all other reversal stages. The dynamic regulation of these isoforms is known to be critical for homeostatic synaptic plasticity [15, 18, 23, 30]; therefore, we hypothesize that the transient dip in circHomer1 expression during chance reversal is likely prompting scaffold rearrangement [18] through Homer1b shuttling within the synapse [31]. However, the inability for circHomer1 expression to return to normal, as in circHomer1 KD animals, would potentially inhibit the synaptic renormalization critical for synapse stabilization, important for learning. The flexibility of synaptic tuning via Homer1 isoform regulation thus appears to be critical during early-mid reversal learning, yet, detrimental during mid-to-late stage reversal. This is similar to the dynamic nature of NMDAR subunits across reversal learning; whereby, the GluN2B subunit, associated with flexibility, is critical during early-to-mid reversal [56] and the GluN2A subunit, associated with stability, dominates during mid-to-late stage reversal [84]. circHomer1 KD animals exhibit a learning deficit during mid-stage (chance) reversal, indicating circHomer1 is likely acting in the transition from plastic to stable synapses. We found a reduction in Grin2a and b isoforms following KD of circHomer1 and differential correlation of these isoforms between circHomer1 KD and control animals, which likely contributes to the behavioral and electrophysiological phenotypes observed.

circHomer1 maintains cellular firing rate in vitro and in vivo

Previous experiments have shown that heightened network activity (induced by bicuculline) results in a robust increase in circHomer1 expression and localization to the dendrites [33]. In this way, circHomer1 appears to act as an immediate early gene similar to its linear counterpart Homer1a, which scales down synaptic weights through scaffold disruption and AMPAR internalization [18, 30]. We have previously demonstrated that circHomer1 KD increases both Homer1b and Homer1a localization to the synapse [31], suggesting a potential compensatory mechanism triggered to maintain balance of postsynaptic firing potential. Previous work indicated circHomer1 expression is inversely correlated with duration of illness [32], which could indicate disease progression reflects downregulation of important synaptic effectors including circHomer1. This is corroborated by our finding that KD of circHomer1 produces long-lasting changes in transcriptional regulation of post-synaptic regulators of glutamatergic neurotransmission. Chronic reduction of circHomer1, as seen in psychiatric and neurodegenerative disease, could potentially disrupt the activity-dependent regulation resulting in sustained downscaling as a homeostatic mechanism to prevent repeated hyperexcitation. If this were the case, we would expect to see the chronic loss of circHomer1 to result in diminished cellular firing rates. Our in vitro experiments indicate reduced baseline neuronal firing frequency following circHomer1 KD, suggesting that reduction of circHomer1 is sufficient to impair spontaneous, AMPAR-dependent, neuronal activity. In vivo, we found that baseline (pre-choice) firing rate is dynamic across reversal learning; however, there is a profound reduction in baseline firing rate of circHomer1 KD neurons during chance reversal, which may be detrimental to learning at this stage and represent an altered homeostatic response to network activity [85]. The reduction in sEPSC frequency in circHomer1 KD neurons may be a result of reduced presynaptic activity through a feedback mechanism or diminished AMPA receptor unsilencing [86] in surrounding postsynaptic terminals resulting from a disruption of Homer1 scaffold integrity.

circHomer1 expression in the OFC is a cognitive-related biomarker

CircHomer1 expression is reduced in multiple patient populations with psychiatric [31, 32] and neurodegenerative disease [38–42]. Its expression is correlated with duration of illness in human patients [32], cognitive performance in rodents during a translational touchscreen task [31], and alterations in neuronal signaling. Importantly, circHomer1 is more stable than linear RNA due to its lack of free ends [37, 87] and its expression in peripheral blood cells correlates with disease [31, 42] and is amenable to antipsychotics [32] indicating its potential role as a predictive biomarker [87] of cognitive dysfunction. While there is a degree of overlap in circHomer1 expression in patients and controls [31, 32] as well as in our animal model [32] (and current work), its expression may be more predictive of a particular behavioral phenotype such as behavioral inflexibility that spans multiple disorders rather than any specific disease as a whole. We present findings that suggest circHomer1 influences neuronal signaling and coordination within the OFC during specific windows critical for efficient learning. We hypothesize based on this work and previous studies that circHomer1 does this by balancing linear Homer1 isoform expression in an activity-dependent manner, and loss of circHomer1 results in unchecked linear isoform regulation required for homeostatic maintenance. The model provided here demonstrates loss of circHomer1 is sufficient to disrupt neural coordination and behavior in otherwise genetically wild-type mice. Alterations in circHomer1 in addition to other genetic risk factors may amplify the phenotypes. Future work will be needed to determine associations with circHomer1 expression and human cognitive flexibility, but our results provide a promising rationale for this work. Because circHomer1 expression is associated with multiple neurological and psychiatric disorders such as Alzheimer’s disease [38–40], Parkinson’s disease [42], mesial temporal lobe epilepsy [41] and SCZ and BD [31, 32], there is likely a convergent mechanism by which synaptic dysfunction results in the ultimate reduction of circHomer1 that promotes brain region-specific phenotypes.

Limitations

Due to the highly specific design needed for an shRNA to target a circRNA backsplice junction [37], we are only able to use a single shRNA. We have previously demonstrated the specificity of this shRNA [31, 32]. Only male mice were used in this study to be consistent with previously published work characterizing circHomer1 KD animals [32]; however, recent work has suggested sex differences may be present regarding loss of circHomer1 [39]. We recognize that the sample size is limited in the behavioral and transcriptional analysis due to experimental design trade-offs whereby we selected a random subset of animals to be used for PCR quantification and chose to use the same animals for behavioral analysis as this was a replication study following our previous work [32]. The molecular and behavioral characteristics of this subset are, however, in accordance with previously published phenotypes [31, 32] demonstrating the specificity of circHomer1 KD within the OFC.

Author contributions

J.L.B, N.M., and A.J.Z. designed the experiments. A.J.Z., J.P.W, and G.P. carried out experiments. A.J.Z wrote the manuscript. J.L.B, N.M., J.P.W., and G.P., edited and revised the manuscript.

Code availability

MATLAB code for analysis of electrophysiological data is available from the corresponding author upon request.

Competing interests

N.M. has a financial interest as co-founder of Circular Genomics Inc. and as inventor of patents related to the use of circRNAs for brain disease diagnostics. G.P. is a current employee for Circular Genomics. N.M. was not involved in data acquisition for this manuscript. A.J.Z., J.P.W. and J.L.B. have no conflicts to disclose.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-024-03188-0.