Prediction error signaling explains neuronal mismatch responses in the medial prefrontal cortex

Abstract

32According to predictive coding theory, perception emerges through the interplay of neural circuits 33 that generate top-down predictions about environmental statistical regularities and those that 34 generate bottom-up error signals to sensory deviations. Prediction error signals are hierarchically 35 organized from subcortical structures to the auditory cortex. Beyond the auditory cortex, the 36 prefrontal cortices integrate error signals to update prediction models. Here, we recorded neuronal 37 activity in the medial prefrontal cortex of the anesthetized rat while presenting oddball and control 38 stimulus sequences, designed to separate prediction errors from repetition suppression effects of 39 mismatch responses. Robust mismatch signals were mostly due to prediction errors. The encoding of 40 a regularity representation and the repetition suppression effect over the course of repeated stimuli 41 were fast. Medial prefrontal cells encode stronger prediction errors than lower levels in the auditory 42 hierarchy. These neurons may, therefore, represent the neuronal basis of a fundamental mechanism 43 of hierarchical inference. 44 45 46 Keywords: auditory processing, predictive coding, prediction error, mismatch negativity, medial 47 prefrontal cortex, anaesthesia, neuronal activity 48 49 3 50 51 100 unpredictable deviations from the auditory background. These cells may, therefore, represent the 101 neuronal basis of predictive activity in FCs. 102 103 104 Results 105Context-dependent responses across fields in mPFC. 106 To seek experimental evidence for predictive coding signals in the mPFC, we recorded neuronal 107 activity across all fields in 33 urethane-anesthetized rats. We recorded 83 multiunits (AGM [medial 108 agranular cortex]: 25; ACC [anterior cingulate cortex]: 20; PL [prelimbic cortex]: 20; IL [infralimbic 109 cortex]: 18) and tested 384 tones (AGM: 13; ACC: 90; PL: 81; IL: 81) as part of oddball paradigms 110 and suitable control sequences, namely, cascade and many-standards conditions (Fig. 1). Figure 2 111 illustrates 5 examples of electrolytic lesions in three Nissl-stained sections at different recording sites 112 across the mPFC. Regardless of their anatomical location in all mPFC fields, we found sound-driven 113 neuronal activity to pure tones (0.6-42.5 kHz; 25-70 dB SPL). We assessed significantly increased 114 responses to sound by comparing baseline-corrected spike counts after stimulus presentation against 115 a simulated null peristimulus time histograms (PSTH) with a firing rate equal to the baseline 116 (detailed in Methods). We tested the neuronal response to each pure tone as either being deviant, 117 standard, or part of a control sequence and only included those frequency tones that demonstrated a 118 significant response to any condition. Thus, 86.2% (331/384) of deviant stimuli evoked a significant 119 neuronal discharge, while 26.6% (102/384) of standard stimuli, 26.6% (102/384) of many-standards 120 stimuli and 32.3% (124/384) of cascade stimuli evoked a firing rate ...

Roving oddball paradigm elicits sensory gating, frequency sensitivity, and long-latency response in common marmosets

“…This relatively slow, positive amplitude component is apparent from approximately 300 to 800 ms post deviant-stimulus-onset, and was not observed in response to other stimuli. This is similar to findings in anaesthetized rodents that display responses to deviant stimuli extending over a comparable time-course ( Casado-Román et al, 2020 , Chen et al, 2015 , O’Reilly, 2019 , O’Reilly and Angsuwatanakul, 2021 , Ruusuvirta et al, 1998 ), and perhaps also to recent observations from conscious humans ( Ruiz-Martínez et al, 2021 ). In urethane-anaesthetized rats, this long-latency activity has been interpreted as evidence for prediction error signaling related to automatic deviance-detection ( Casado-Román et al, 2020 ).…”

“…This is similar to findings in anaesthetized rodents that display responses to deviant stimuli extending over a comparable time-course ( Casado-Román et al, 2020 , Chen et al, 2015 , O’Reilly, 2019 , O’Reilly and Angsuwatanakul, 2021 , Ruusuvirta et al, 1998 ), and perhaps also to recent observations from conscious humans ( Ruiz-Martínez et al, 2021 ). In urethane-anaesthetized rats, this long-latency activity has been interpreted as evidence for prediction error signaling related to automatic deviance-detection ( Casado-Román et al, 2020 ). In urethane-anaesthetized mice these waveform features have also been reported to satisfy some of the requirements for a "genuine" MMN-like response ( Harms et al, 2016 , O’Reilly, 2019 , O’Reilly and Angsuwatanakul, 2021 ); although they were also elicited by deviant-alone control paradigm stimuli, and not by lower intensity oddballs ( Casado-Román et al, 2020 , O’Reilly, 2019 ).…”

“…In urethane-anaesthetized rats, this long-latency activity has been interpreted as evidence for prediction error signaling related to automatic deviance-detection ( Casado-Román et al, 2020 ). In urethane-anaesthetized mice these waveform features have also been reported to satisfy some of the requirements for a "genuine" MMN-like response ( Harms et al, 2016 , O’Reilly, 2019 , O’Reilly and Angsuwatanakul, 2021 ); although they were also elicited by deviant-alone control paradigm stimuli, and not by lower intensity oddballs ( Casado-Román et al, 2020 , O’Reilly, 2019 ). Long-latency single-unit spikes triggered after target stimuli in a behavioral task have also recently been observed from the inferior colliculus of awake marmosets, indicating that this long-latency MMN-like response could involve subcortical communication ( Shaheen et al, 2021 ).…”

“…There is evidence from both animals and humans that deviance-detection may occur later than previously considered ( Casado-Román et al, 2020 , Chen et al, 2015 , O’Reilly, 2019 , O’Reilly and Angsuwatanakul, 2021 , Ruiz-Martínez et al, 2021 ). As such, the long-latency response to deviant stimuli was extracted using a double-epoch waveform subtraction ( O’Reilly, 2019 ).…”

Cortico-Fugal Regulation of Predictive Coding

“…We compared the responses of adapting and facilitating neurons in both central and shell regions of IC to tones in the cascade versus the many standards context (Figure 3 -Figure Supplement 2A). We found no significant differences in firing rates to the cascade versus the many standards contexts (Figure 3 -Figure Supplement 2B, C, Table 1), suggesting that the global structure of the cascade sequence does not significantly affect how neurons in IC respond to this stimulus, as has been shown in other structures (Casado-Román et al, 2020;Parras et al, 2021).…”