Bulk medial prefrontal cortex activity comports with a functional bipartition between salience detection and movement gain
Curation statements for this article:-
Curated by eLife
eLife Assessment
This useful study asks whether medial prefrontal cortex activity reflects prediction error or instead tracks salience and movement across appetitive and aversive tasks. The movement-related findings are the supported by convincing evidence, but the evidence for the other conclusions is incomplete. The design confounds prediction-error sign with outcome delivery versus omission, bulk photometry cannot rule out prediction-error coding, and opsin is confounded with sex in the optogenetic experiments. This study will be of interest to those interested in medial prefrontal cortex, learning, and related processes.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Reward prediction error (RPE) is a key function putatively realized by brain-wide neural states to drive learning and adaptive responding. The medial prefrontal cortex (mPFC) is among the many areas canonically implicated in RPE computation. Yet, whether mPFC RPE processing differs as a function of valence, value, and sign is not well-understood. Here, we performed in vivo mPFC calcium imaging in male and female rats engaged in an appetitive or aversive Pavlovian task designed to elicit RPE, along with continuous, unbiased behavioral monitoring. We found that short-latency bulk mPFC activity reports salience independent of valence, value, or RPE. Surprisingly, as subsequently validated with in vivo bidirectional optogenetics, we show that salience-null mPFC activity modulates generalized movement in a valence- and task-agnostic fashion. Together, these results challenge the pervasive notion of a unitary, region-wide representation of cognitive processes, including RPE signaling, by bulk mPFC activity, and instead, support a functional bipartition schema combining reactive detection of integrated salience and dynamic modulation of movement gain.
Article activity feed
-
eLife Assessment
This useful study asks whether medial prefrontal cortex activity reflects prediction error or instead tracks salience and movement across appetitive and aversive tasks. The movement-related findings are the supported by convincing evidence, but the evidence for the other conclusions is incomplete. The design confounds prediction-error sign with outcome delivery versus omission, bulk photometry cannot rule out prediction-error coding, and opsin is confounded with sex in the optogenetic experiments. This study will be of interest to those interested in medial prefrontal cortex, learning, and related processes.
-
Reviewer #1 (Public review):
This study investigates whether mPFC activity reflects prediction error or salience across appetitive and aversive tasks. The authors recorded bulk calcium activity and used optogenetic manipulations to test its relationship to behavior. The cross-valence comparison, combination of photometry and optogenetic manipulations, and large cohorts are strengths. However, the experimental designs do not clearly distinguish prediction error from salience, and the optogenetic findings are difficult to interpret because the manipulations are confounded with sex. Thus, although the study addresses an important question and its methods and data may be useful to the field, the evidence does not establish that mPFC codes salience or fully support the authors' conclusions. Below are my main concerns.
(1) The authors find no …
Reviewer #1 (Public review):
This study investigates whether mPFC activity reflects prediction error or salience across appetitive and aversive tasks. The authors recorded bulk calcium activity and used optogenetic manipulations to test its relationship to behavior. The cross-valence comparison, combination of photometry and optogenetic manipulations, and large cohorts are strengths. However, the experimental designs do not clearly distinguish prediction error from salience, and the optogenetic findings are difficult to interpret because the manipulations are confounded with sex. Thus, although the study addresses an important question and its methods and data may be useful to the field, the evidence does not establish that mPFC codes salience or fully support the authors' conclusions. Below are my main concerns.
(1) The authors find no evidence of prediction-error signaling in mPFC and argue that salience explains their findings. I have concerns about this interpretation. Calcium responses could reflect sensory-event detection rather than salience. The stronger-shock manipulation is informative because mild and strong shocks evoke similar mPFC responses despite their different physical intensities. This argues against a simple relationship between the signal and stimulus magnitude. However, if salience drives neural responses, it is difficult to explain why doubling shock intensity did not increase bulk calcium activity: "the magnitude of increase did not scale with the magnitude of outcome value or RPE" (line 491, Figure 2f). A stronger shock is more painful, more unexpected, and more salient than a mild one.
(2) The omission trials are also difficult to interpret. The absence of calcium signals during shock or pellet omission cannot distinguish between a lack of prediction error, the absence of a sensory event at the expected outcome time, or insufficient photometry sensitivity. If salience governs mPFC responses, the absence of shock may itself be the most salient outcome for the animal because it is surprising.
(3) The conclusions overstate what fiber photometry can show. Bulk signals can average together neurons with opposing response signs. The authors can conclude that bulk mPFC photometry did not distinguish the tested outcome conditions in a manner consistent with signed prediction error. They cannot conclude from these data that mPFC lacks prediction-error or value coding.
(4) The movement correlation is interesting, but the authors should discuss prior work linking prefrontal activity to freezing behavior (Burgos-Robles et al., 2009; Kyriazi et al., 2020; Casanova et al., 2024). Additionally, cross-correlations are computed on event-averaged traces, which show shared event-locking, not continuous coupling.
(5) The manuscript also states that "fluctuations in Ca2+ signal were fully explained by integrated salience and general uninstructed movement across all stages of learning regardless of task-defined behavioral performance" (lines 868-870). The authors do not report a variance-explained model to support this statement.
(6) Because the central claim is that bulk mPFC activity tracks movement, the authors should assess whether movement-related artifacts contribute to the signal, particularly during vigorous movements such as darting and shock responses. They used 405 nm light as an isosbestic control for calcium-independent motion artifacts (lines 176-178) and corrected the GCaMP signal by fitting the 405 nm signal to the 465 nm signal before calculating ΔF/F (lines 184-187; Supplemental Figure 1a-c). This correction may not remove all artifacts, including hemodynamic changes that affect the two wavelengths differently. To help readers assess whether event-related changes in the calcium signal remain after correction, the authors should show the raw 405 nm and 465 nm signals, as well as the corrected GCaMP signal, aligned to darting onset, freezing onset and offset, and shock delivery.
(7) The freezing/darting dissociation is interesting but confounded. Darting is defined differently across cohorts: by a trained classifier using a qualitative definition ("sudden and rapid escape-like movement") in the photometry cohort, and by a velocity threshold ({greater than or equal to}54 cm/s, {greater than or equal to}0.2 s) in the optogenetic cohort. The authors should provide a quantitative definition for both cohorts and state whether shock-elicited escape responses were excluded.
(8) The classification of rats as "pellet-responsive" or "non-responsive" is problematic. First, rats are split at zero on a distribution of post-pellet Ca2+ rises that appears unimodal (Figure 6a), and the two groups are then compared on the same calcium measure used to define them. The representational similarity analysis does not validate the grouping independently, because it uses the same pellet-evoked responses. Responsive rats also show larger responses to every task event, not only to the pellet, which raises the possibility that the groups differ in signal quality, expression, or fiber placement. The authors should compare signal-to-noise ratio, expression, and implant location between groups, and treat pellet responsiveness as a continuous variable instead of dividing rats into two groups.
Second, the behavioral differences appeared only after a change of analysis. Summary statistics during probe testing showed no differences in reward seeking or velocity between groups (Figure 6g,h; Supplemental Figure 6). The authors then applied functional linear mixed models to the training data and reported lower reward seeking during the cue and delay, but greater reward seeking immediately after pellet delivery, in responsive rats (Figure 6i). The authors interpret these results as a more precise retrieval strategy, but the direction could be reversed. Rats that approach the magazine promptly after pellet delivery would be expected to show a calcium increase linked to that action, which, by the authors' own account, reflects movement. The authors should show, on a trial-by-trial basis, that the pellet-evoked calcium rise precedes rather than follows magazine approach.
(9) I have several concerns about the optogenetic experiment. The constructs are not equivalent. The opsins are expressed under different promoters (Syn for ChR2, nEF for NpHR), and the GFP control uses a different serotype (AAV2 vs AAV8), so viral spread and cell-type expression may differ between groups. Both opsins are pan-neuronal and therefore also target interneurons, making the net effect of stimulation or inhibition difficult to predict. The authors should show expression spread for each construct and discuss this issue.
(10) The opsin is confounded with sex. Figure 4b shows that most darters are female. However, darting is also present in males, and in the optogenetic experiment, where channelrhodopsin (ChR2) was expressed in males and halorhodopsin (NpHR) in females, the authors report darters in all groups (line 688: "During baseline testing, 5 out of 8 ChR2 rats, 2 out of 7 NpHR rats, and 3 out of 6 GFP rats were identified as darters"). Because distinct opsins were injected in females and males, the effects of opsin and sex cannot be separated.
(11) Several optogenetic conclusions rest on within-group comparisons rather than comparisons with GFP controls. During the intertrial interval, GFP rats showed the same increase in freezing and decrease in velocity during light delivery as NpHR rats (Figure 8a,b), which the authors attribute to freezing naturally increasing as the next trial approaches (lines 724-726). In the context-control session (Figure 8c,d), the significant interaction is again followed only by within-group pre-versus-ON comparisons. The authors should test directly whether light-induced changes in ChR2 and NpHR rats differ from those in GFP rats. During the task, summary statistics showed no effect of inhibition (Figure 7f-i). The delay-period reduction in movement and darting (Figure 7h) is a main effect of laser across all groups, including GFP, yet the Discussion states that manipulation "in either direction decreased movement velocity (i.e., suppressed darting)" (line 813), while the Results describe the ChR2 effect as nonsignificant (line 692). Given that opsin is confounded with sex (point 10), these effects should be interpreted cautiously.
(12) The authors hypothesized that stimulation and inhibition would have opposite effects on darting, yet both reduced darting. Still, the Discussion states that the hypothesis was "largely validated" (line 815).
(13) The mPFC is composed of PL and IL, which have opposite effects in fear conditioning, and the recordings may also include dorsal regions. The injection coordinates (ML +1.4; lines 116-117) and the large volume of AAV9 (500 nL; line 119) make it likely that expression extended into the anterior cingulate and secondary motor cortex. Fiber placements also appear more dispersed in males than in females (Figure 1c). A movement-related signal could therefore originate in motor or cingulate regions rather than in PL, and the same concern applies to opsin spread in the optogenetic experiments. I would like to see expression maps and the movement correlation analyzed by fiber placement. In addition, the atlas levels shown in Figure 1c (+5.64 to +4.20) do not match the stated AP coordinate (+3.2) (line 116).
(14) The criteria for learning are not defined. The Methods state only that "learning was indicated by an increase" in conditioned freezing (line 147) or in anticipatory reward seeking (line 162), without specifying a threshold or performance criterion. The two tasks also differ substantially in training. The aversive group received a single 10-trial conditioning session, whereas the appetitive group received 6-9 sessions of 20 trials, and it is not stated what determined the number of sessions for each rat. As a result, the two groups may have been at different stages of learning when prediction-error probe testing began, which complicates the cross-valence comparisons. The authors should state their learning criterion, report whether any animals failed to meet it and were excluded, show individual learning curves for both tasks, and explain how the number of appetitive training sessions was determined.
(15) A null result in a bulk signal cannot rule out prediction-error coding, because intermixed positive- and negative-coding neurons could cancel each other out (point 3). The passages on the "ontological validity" of dopamine prediction error and on mammalian brain evolution are also not connected to the data.
-
Reviewer #2 (Public review):
Summary:
Hou and colleagues aimed to characterize bulk medial prefrontal cortex calcium signal surrounding reward prediction error processing to determine encoding of value, valence, and sign. While a salience signal was observed, they also identified a causal role for mPFC activity surrounding movement, independent of task.
Strengths:
Overall, this manuscript from Hou and colleagues is an enjoyable read and a well-considered manuscript presenting findings that clearly were different from hypothesized. I applaud the authors for their deep dive into this and the push toward more causal interrogation of the mPFC in regulating movement-related outcomes, and think as a whole the manuscript adds to the literature. Major strengths include the careful consideration of appetitive vs. aversive behaviors and …
Reviewer #2 (Public review):
Summary:
Hou and colleagues aimed to characterize bulk medial prefrontal cortex calcium signal surrounding reward prediction error processing to determine encoding of value, valence, and sign. While a salience signal was observed, they also identified a causal role for mPFC activity surrounding movement, independent of task.
Strengths:
Overall, this manuscript from Hou and colleagues is an enjoyable read and a well-considered manuscript presenting findings that clearly were different from hypothesized. I applaud the authors for their deep dive into this and the push toward more causal interrogation of the mPFC in regulating movement-related outcomes, and think as a whole the manuscript adds to the literature. Major strengths include the careful consideration of appetitive vs. aversive behaviors and leveraging of individual differences, the integration of fiber photometry data with optogenetic manipulations, and the cautious interpretation of findings.
Weaknesses:
Some minor weaknesses were noted, including a lack of clarity/consideration of the specific medial prefrontal cortex subregion targeted (appears to be prelimbic), which has important implications for generalizability of the findings due to downstream connectivity, and potentially also integration with what is known for other regions (e.g., ACC) in regulating behavior. In addition, while the authors' conclusions regarding bulk signal are sound within the dataset, and their conclusion to interpret these signals with caution is valid, it may be beneficial to also interpret these findings in light of other literature describing mPFC subregion activity in the regulation of movement and actions broadly, including areas where there are alignment vs disagreement.
-
Reviewer #3 (Public review):
Hou and colleagues used fiber photometry to record bulk calcium activity in the medial prefrontal cortex (mPFC) of rats trained in appetitive (pellet) and aversive (foot shock) cue-outcome tasks. Probe sessions included omitted and larger-than-expected outcomes to test for prediction error signaling. They related the calcium signal to freezing, darting, and reward seeking, and they used optogenetic excitation (ChR2) and inhibition (NpHR) of the mPFC to test its causal role in behavior. The authors conclude that bulk mPFC activity does not encode signed reward prediction error. They argue instead that it reflects perceptual salience combined with uninstructed movement, and that manipulating mPFC activity alters movement, including darting. I agree that the movement findings are the best supported part of the …
Reviewer #3 (Public review):
Hou and colleagues used fiber photometry to record bulk calcium activity in the medial prefrontal cortex (mPFC) of rats trained in appetitive (pellet) and aversive (foot shock) cue-outcome tasks. Probe sessions included omitted and larger-than-expected outcomes to test for prediction error signaling. They related the calcium signal to freezing, darting, and reward seeking, and they used optogenetic excitation (ChR2) and inhibition (NpHR) of the mPFC to test its causal role in behavior. The authors conclude that bulk mPFC activity does not encode signed reward prediction error. They argue instead that it reflects perceptual salience combined with uninstructed movement, and that manipulating mPFC activity alters movement, including darting. I agree that the movement findings are the best supported part of the work. However, I have concerns about the prediction error design, in which error sign is confounded with outcome delivery versus omission. I am also concerned that the salience account is not clearly defined and is not well supported by the data, that opsin is confounded with sex in the optogenetic experiments, and that omnibus statistics are not reported and the figures are hard to read. Finally, I have broader reservations about how much bulk photometry in this design can tell us about what the mPFC encodes.
(1) Their prediction error logic for foot shock is backwards from how it is normally treated. They define RPE as value-signed (better than expected = positive, worse = negative), so shock omission is a positive error and a stronger-than-expected foot shock is a negative error. But with shock as the US, omission should produce a negative error, while presenting a stronger-than-expected foot shock should produce a positive error. More of a problem is that, with their design, the sign is confounded with whether a stimulus is delivered or omitted in both tasks. So the lack of a bidirectional RPE signal mostly comes down to omissions not producing a response and deliveries producing one.
(2) Many important graphs contain no y-axis and only an indicator bar. This is true for all of the time course plots (PSTHs), which also have no zero line. This makes the graphs very frustrating to read.
(3) I am not sure what they mean by perceptual salience, as it is never defined when it is used in the results. There is a definition in the discussion, but no predictions are laid out ahead of time. They report that the lack of an mPFC response to shock omission, coupled with identical responding to mild and strong foot shock, is consistent with perceptual salience. This seems circular to me, and it is not clear what this finding actually means. If anything, a perceptual salience account would predict a bigger response to the stronger shock and the larger reward, which they don't find. And by their own definition (which includes surprise) an unexpected omission should be salient too. Similarly, I didn't follow the reward analysis that was also argued to reflect perceptual salience, especially since only some of the rats (the pellet responsive ones) even show a pellet response.
(4) The authors' conclusions about movement appear to be better supported by the data. Although, again, the graphs were difficult to follow, and it took me quite a while to figure out what the snowflake and the dartboard icons were referring to, as they are not defined in the legends. For the optogenetics, ChR2 was only given to males and NpHR only to females, so opsin and sex are confounded. Also, both ChR2 and NpHR reduced movement after cue offset, so calling the effect bidirectional seems like a stretch outside of the context session.
(5) There are no statistics presented in the results section to support any claims. Significance indicators are given in the figure captions, and some analyses are explained there. But often the appropriate overall tests are not reported. For example, Figure 2 shows their first behavior and calcium imaging time course data. I would have expected first some kind of analysis like repeated measures ANOVA to reveal the main effects and interactions for time, trial, condition, etc. None of that is performed or is at least not reported (maybe it is in Supplemental Table 1). Instead, the time courses are reduced to slopes in chosen windows and compared with lots of separate paired t tests, and it is not explained why those windows were picked. The individual tests look to support the patterns, but they are more convincing when significant main effects or interactions of interest are reported.
Overall takeaway:
After reading this, and even setting aside the concerns above, my main reservation is about the overall contribution. To me, the manuscript reads as "bulk fiber photometry in this prep shows mPFC signals related to movement." I think that is probably true, and the movement data are the strongest part of the paper. But I'm not sure it moves the needle much on what we think the mPFC does. If one were to set out to rigorously test movement coding against these alternatives, I think it would take two things. First, the behavior would need to be designed to decorrelate movement from the other variables. That could mean cues or outcomes that differ in value or prediction error while eliciting matched movement, or the same outcome reached through different movement requirements. Second, the recordings would need single-unit or cellular-resolution precision. Bulk fiber sums activity across the whole population under the fiber, so it can't tell a uniform movement signal apart from a mix of neurons that separately encode movement, salience, or prediction error. Opposing signals could also cancel out in the bulk average. For example, prediction error neurons with opposite signs could sum to nothing, which may help explain why they don't see one.
-
-
RRIDs were included in this paper. We value the author's support of reproducibility. #STMpublishing #RRID #OpenScience
-
Functional bipartition of medial prefrontal cortex into salience detection and movement gain https://www.biorxiv.org/content/10.64898/2026.03.18.712525v1
-
Functional bipartition of medial prefrontal cortex into salience detection and movement gain https://www.biorxiv.org/content/10.64898/2026.03.18.712525v1
-