DANDI:000462

Mice on average ran 34 ± 2 (mean ± 95% CI) laps in the Rewarded condition, at which point, reward was turned off and imaging in the Unrewarded environment continued (28 ± 4 laps)

Methods, Imaging sessions, paragraph 1

Complete laps (start in the first 10 cm, reach the last 5 cm) average 35.3 +/- 3.7 in R and 29.7 +/- 9.2 in UR over 10 mice, 28.2 +/- 10.2 in RR and 32.8 +/- 16.3 in the novel session over 5, all within tolerance. Counting partial laps gives 36.8 in R (outside) and 30.8, 29.6 and 34.2. Dropping the frames the paper's code drops (task_framestokeep, which keeps only the first 11100 of 20000 R frames of NR6) gives 33.2, 29.1, 27.6 and 32.8.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).

We found that the decoder was able to accurately predict position on the final laps in R and initial laps in UR before abruptly underperforming

Results, Lowering reward expectation changes spatial encoding, paragraph 2; Figure 1d

With the Methods decoder (first 60% of R laps, circular track) the first 3 UR laps are within 0.1 of the held-out R laps and RE low laps are below both in 8 of 10 mice under L1, L2 and L3. The criterion asks for 9 of 12 mice (75%); applied as the same fraction of the 10 mice here it needs ceil(0.75 x 10) = 8, which holds; a literal 9 of 10 does not. The code's decoder (first 50% of R frames) gives 6 of 10, and the decoder without the circular treatment 2 of 10.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The criterion's 9 of 12 mice is applied as the same fraction, 75%, of the 10 mice in the dandiset, at least 8 (ceil(0.75 x 10)); a literal 9 of 10 is not reached (shown).
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).

On average, pre-licking continued for a few laps before rapidly dropping, and interestingly, decoder performance sharply dropped around the same lap when pre-licking reached zero

Results, Lowering reward expectation changes spatial encoding, paragraph 2; Figure 1e; Methods, Decoder performance with different behavioral parameters, Licking behavior

The reverse Boltzmann fit to the mean of the 3-lap smoothed UR R2 over 10 mice (first 15 laps) has r = 0.979 and R2 = 0.959 (paper 0.94); the first zero of its second derivative is at lap 11.23, rounded to 11 (paper 10 +/- 1); the smoothed lick curve reaches its minimum at lap 14, 3 laps away, within the 3 laps allowed.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.

Interestingly, in RE low , decoder error had increased at all locations across the track (Fig. 1g ), and not just around the reward site

Results, Lowering reward expectation changes spatial encoding, paragraph 4; Figure 1g

What the paper reports

• R against RE low decoder error at each of start, middle and end: two-sided paired t test, p < 0.01 (n = 12 mice)

Under L1 the mean decoder error in RE low against R is 17.1 against 4.9 cm at the start, 33.9 against 5.6 cm in the middle and 17.3 against 4.4 cm at the end, paired t p < 0.0001, 0.0001 and < 0.0001 over 10 mice; 3 of 3 groups under every lick stop reading and with the code's decoder. In the 5 RR mice the errors (12.1, 21.0, 11.6 cm) lie between R and RE low. The position groups are those of the Figure 5d legend, as the criterion reads them.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).

We found that if we instead used different criteria to identify when licking had stopped, i.e., the first lap with no licks, or 4 consecutive laps with no licks, our results were unaffected. ... However, with 6 consecutive laps with no licks, our results differed

Methods, Decoder performance with different behavioral parameters, Licking behavior; Supplementary Fig. 4

With RE low starting at the first lap of each run of no-lick laps, RE low R2 is below RE high for the first no-lick lap (Bonferroni p = 0.006), 2 and 4 laps (0.012; the same split, as in every mouse the first two consecutive no-lick laps begin a run of 6 or more) and the grouping by licks (0.007); RE high medians 0.910 to 0.914 and RE low 0.661 to 0.671 are inside the widened ranges; median R 0.976 and RR 0.842 are within tolerance. Starting RE low at the Nth lap or after the run, the 4-lap test is not significant after correction (0.066, 0.065). The 6-lap definition gives Bonferroni p = 0.30 and 0.12 under those two readings, reported, not judged.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.

Importantly, we found a similar reduction in decoder performance in UR when using only RE low engaged laps (Supplementary Fig. 6B , Mean decoder R 2 [95% CI]: ... RE low : Engaged = 0.65 [0.49 0.81]), although disengaged laps did further reduce decoder performance (RE low : Disengaged = 0.31 [0.06 0.5])

Results, Changes in spatial encoding associated with diminished reward expectation are not due to disengagement, paragraph 1; Supplementary Fig. 6b

What the paper reports

• RE low engaged laps: 95% CI 0.49 to 0.81 (n = 12 mice)
• RE low disengaged laps: 95% CI 0.06 to 0.5 (n = 12 mice)

Main engagement split, L1, 10 mice - engaged RE low laps decode at R2 0.759 [0.568 0.950] (paper 0.65 [0.49 0.81]) and disengaged laps at 0.424 [0.214 0.634] (paper 0.31 [0.06 0.5]), both within tolerance; engaged laps are below R (Wilcoxon p = 0.0020) and disengaged laps below engaged on average and in 9 of 10 mice. L2 and L3 give 0.696 and 0.658 (engaged) and 0.448 and 0.444 (disengaged), also within tolerance, with R against engaged p = 0.0039 over the 9 mice with engaged laps. The code's ratio (variant) gives 0.626 and 0.473 under L1. Some per-mouse means rest on 1 to 5 laps (NR21, NR32, NR34, CFC4, CFC16 for engaged laps under L1).

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).

The probability of disengaged laps in the first half of RE low was similar to the second half ( P = 0.14). Furthermore, the distribution of disengaged laps throughout the session did not differ from a uniform distribution ( P = 0.7)

Results, Changes in spatial encoding associated with diminished reward expectation are not due to disengagement, paragraph 1

What the paper reports

• first against second half of RE low, probability of disengaged laps: p = 0.14 (n = 12 mice)
• distribution of disengaged laps against uniform: p = 0.7

Main engagement split. Under L1 the fraction of disengaged laps is 0.383 in the first and 0.541 in the second half of RE low (10 mice; Shapiro-Wilk p = 0.086, paired t p = 0.094, Wilcoxon p = 0.094), and the positions of the 125 disengaged laps do not differ from uniform (KS D = 0.053, p = 0.853). L2 and L3 give p from 0.078 to 0.975. Both tests give p >= 0.05 under every reading, as the criterion asks for a claim of absence, though the second half has 0.16 to 0.17 more disengaged laps under L1 and L2. The code's ratio (variant) gives 0.284 and 0.324 (paired t p = 0.639, KS p = 0.541).

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).

place field trial-to-trial reliability 23 (Rewarded = 0.47 [0.38 0.57], RE low : Engaged = 0.32 [0.15 0.48])

Results, Place fields show diminished reliability, paragraph 1; Figure 3b

What the paper reports

• R: 95% CI 0.38 to 0.57 (n = 3 mice, 605 fields)
• RE low engaged: 95% CI 0.15 to 0.48 (n = 3 mice, 663 fields)
• RR: 95% CI 0.31 to 0.51 (n = 3 mice, 674 fields)

Per-mouse means of the 2 qualifying mice under L1 - R 0.545, RE low 0.381, RR 0.495, within 0.47 +/- 0.10, 0.32 +/- 0.17 and 0.41 +/- 0.10; pooled fields 0.554, 0.351, 0.478, also within. RE low is below R (pooled fields, unpaired t p = 2.0e-26; KS p = 2.4e-20). L2 and L3 also hold. Under L1c (4 mice) R (0.570) and RR (0.520) are just outside.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• Transients outside the field are removed as whole transients (runs of significant frames) that do not touch the field's bins, following the Methods (Place field parameters); the reliability is the mean of the off-diagonal lap by lap correlations, laps without activity left out, times the fraction of laps with a transient in the field.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

Lastly, we found DCZ and CNO tended to induce a reduction in place field over-representation of the reward site compared to saline controls, but this did not reach significance (Fig. 4m ; P = 0.06) and the effect size was less than RE low

Results, Bilateral VTA inhibition largely replicates lowering reward expectation, paragraph 5; Figure 4k to m; Supplementary Fig. 17e; Methods, Reward over-representation

What the paper reports

• change in end minus middle place field percentage, DCZ/CNO against saline: two-sided paired t test, p = 0.06

Change in end minus middle percentage (after minus before), 4 mice - saline +4.3, DCZ +2.1, CNO +5.8; DCZ against saline paired t p = 0.402 (not significant, as the claim says), and DCZ rises less than saline. RE low against R (L1, engaged laps) -11.9 over the 3 mice with enough engaged laps (009 has 2), larger in size than DCZ; under L3 engaged laps it is -1.1, smaller. CNO against saline Wilcoxon p = 0.625; CNO rises more than saline, which the criterion does not judge (asked of a reader). Gaussian means 108.5 to 115.0 cm, all within 10% of the paper's.

How the test differed from the paper’s method

• The dandiset holds 4 of the 6 DREADD mice, all 4 with saline (paper 5).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The DREADD offset is removed as for fig4f-dreadd-spatial-correlation. Assumption.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.
• Behaviour of sub-014 from BeforeCNO on is realigned by 6000 frames (found in the notebook).

Indeed, decoder accuracy in RE high laps was similar to R and was significantly lower in RE low laps (Fig. 1f , mean decoder R 2 [95% confidence intervals (CIs)] ... R = 0.95 [0.93 0.97], RE high = 0.90 [0.87 0.93], RE low = 0.65 [0.54 0.75]). ... The decoder accuracy somewhat recovered following reward re-introduction in RR laps, although it remained lower than in R (RR = 0.82 [0.77 0.87])

Results, Lowering reward expectation changes spatial encoding, paragraph 3; Figure 1f

What the paper reports

• R, mean decoder R2 over mice: 95% CI 0.93 to 0.97 (n = 12 mice)
• RE high: 95% CI 0.87 to 0.93 (n = 12 mice)
• RE low: 95% CI 0.54 to 0.75 (n = 12 mice)
• RR: 95% CI 0.77 to 0.87 (n = 6 mice)

Under L1 the per-mouse means (code's lap draws, computed as their expectation) are R 0.960 [0.926 0.994], RE high 0.901 [0.863 0.939], RE low 0.606 [0.442 0.769] (10 mice) and RR 0.861 [0.771 0.951] (5 mice), all within tolerance; L2 also, L3 not (RE high 0.868). R against RE low p = 0.0008 (Bonferroni 0.0048); R against RR p = 0.0345 (Bonferroni 0.207); R against RE high p = 0.0047 (Bonferroni 0.028), so RE high is not similar to R as the criterion requires (p >= 0.05). The code's decoder gives the same pattern (R 0.954, RE high 0.899, RE low 0.601, RR 0.851; R against RE high p = 0.0073). Over all tested R laps R is 0.977, outside 0.95 +/- 0.02.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.

Decoder R2 for each lap in each mouse (circles) in different conditions (Number of laps: R = 399, REhigh = 90, RElow = 244, RR = 121). Mean ! standard deviation is shown alongside (Mean (standard deviation): R = 0.97 (0.046), REhigh = 0.86 (0.14), RElow = 0.56 (0.34), RR = 0.79 (0.19)

Supplementary Fig. 4f

Under L1, pooled over 10 mice, R has 353 complete laps (mean R2 0.990, SD 0.039, training laps included; held-out laps 145, 0.977), RE high 66 laps (0.888, SD 0.144), RE low 231 (0.646, SD 0.436), RR 141 (0.869, SD 0.155). Every mean is within the stated SD under every reading. The lap counts are not within tolerance under any reading - R 353 against 399 +/- 40, RE high 66 to 86 against 90 +/- 9 with RE low 231 to 211 against 244 +/- 24, RR 141 against 121 +/- 12. Two mice are missing. RR has 141 laps from 5 mice against the paper's 121 from 6, so the missing mice do not explain that count.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.

70% of the laps in UR had lap speeds that matched lap speeds in R (matched speed laps = 233, slower speed laps = 101)

Supplementary Fig. 5; Methods, Time taken to complete a lap

Matching UR laps to the range of R lap times per mouse (minimum to maximum) gives 172 matched and 125 slower of 297 laps (58%; paper 233 and 101, 70%); lap times 12.23 s in R, 13.38 s matched and 28.19 s slower (paper 7.34, 7.4, 20). The median over mice of the per-mouse R2 is R 0.985, matched 0.763 and slower 0.695, within tolerance, and matched laps are below R (Wilcoxon p = 0.0059 with Bonferroni). The 5th to 95th percentile reading gives 118 and 179 laps; counting running time only, as the code does, gives 203 and 94 laps and 8.50, 9.16 and 15.03 s. The clock check in the notebook shows 46 to 49 frames (1.53 to 1.63 s at 30 Hz) from reward onset to teleport, against the 1.5 s pause of the Methods, and 3.36 to 4.8 R laps per minute from the lap counts and condition lengths, so the longer lap times are not a clock error; lap time does not depend on the volts-to-cm scale.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm. Assumption. The 30 Hz clock agrees with the 1.5 s pause after the reward (46 to 49 frames from reward onset to teleport in every rewarded condition), and lap time does not depend on the scale.
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.

We found RE low caused partial remapping as shown by a drop in the spatial correlation of place fields between R and UR conditions (using only RE low : Engaged laps; Fig. 2b–d )

Results, Lowering reward expectation induces place cell remapping, paragraph 2; Figure 2b, 2c

Reading S2 (first against second half of R in the same mice), 2 qualifying mice, 238 R place cells with every correlation defined under L1 - R against RE low 0.503, within R 0.799 (paired t p = 1.2e-28), R against RR 0.579, above RE low (p = 8.5e-04) and below within R; L2 and L3 the same, and the L1c variant with 4 mice (0.556, 0.857, 0.678). Reading S1 needs the control mice, which are not in the dandiset; the criterion requires S1 and S2 to agree, so the test can be run only in part.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• Reading S1 (within-R from the control mice) could not be run, as the control mice are not in the dandiset; the criterion asks for S1 and S2 to agree, so the outcome can be partial at most.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

Interestingly, partial remapping caused by RE low occurred at all locations throughout the environment, and not just at locations near the reward site (Fig. 2d )

Results, Lowering reward expectation induces place cell remapping, paragraph 2; Figure 2d

What the paper reports

• R against RE low spatial correlation at each of start, middle and end: two-sided paired t test, p < 0.01 (n = 605 place cells defined in R, 3 mice)

Reading S2, L1, 2 qualifying mice - R against RE low below within R at the start (0.514 against 0.820, n = 53, p = 2.5e-07), middle (0.478 against 0.801, n = 124, p = 3.9e-17) and end (0.540 against 0.777, n = 62, p = 2.6e-07); 3 of 3 groups under every lick stop reading and under L1c. Reading S1, which must agree, cannot be run (no control mice).

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• Reading S2 only, as for fig2c-spatial-correlation-drop; S1 cannot be run, so the outcome can be partial at most.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

This demonstrates that diminished reward expectation leads to a spatial code in CA1 with low quality place fields at all locations

Results, Place fields show diminished reliability, paragraph 1; Figure 3c

What the paper reports

• R against RE low parameters at each of start, middle and end: two tailed KS-test, p < 0.01

Under L1, 2 qualifying mice - reliability lower in RE low at the start (0.257 against 0.529, KS p = 2.8e-09), middle (p = 2.5e-09) and end (p = 4.7e-07); out/in higher in the middle (0.085 against 0.065, p = 2.7e-03) but not at the start (p = 0.22) or end (p = 0.013). Three groups for one measure, the criterion's partial case; L2 and L3 give the same, and L1c 3 and 2 groups.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

We found that the spatial decoding and place cell parameters were poor in the initial laps of the novel environment, however, with time, as the animal learned the environment and developed reward expectation (Supplementary Fig. 13B ), these properties became better and comparable to properties in R and RR

Results, Place fields show diminished reliability, paragraph 2; Supplementary Fig. 13

Five of the 6 novel mice. Late laps (last 12) against early laps (first 12) - decoder R2 0.919 against 0.821 (paired t p = 0.0093), reliability of block fields 0.392 against 0.320 (p = 0.038) and of session fields measured on the blocks 0.434 against 0.230 (p = 0.0036), out/in ratio 0.080 against 0.091 (p = 0.34) and 0.203 against 0.550 (p = 0.096), pre-reward licks in the zone per lap 3.33 against 1.78 (p = 0.39). Two of the 4 tests hold under both readings of the block parameters. The session-field out/in means are pulled up by fields with almost no activity in a 12-lap block (up to 20 per mouse above 1); the medians are 0.121 to 0.158 early and 0.040 to 0.083 late. Within-condition R2 (sliding 20% test block, down-sampled to the fewest laps) is 0.907 to 0.989 in R, 0.870 to 0.974 in RR and 0.817 to 0.941 in N.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The decoder takes each prediction at its nearest copy of the true bin before R2 and error, as the criterion's 'track treated as circular' (Methods, Position decoding) implies; without it R2 is far lower (shown).
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.
• Pre-licking is counted as licks in the reward zone before the reward is delivered; the paper does not say how it measured pre-licking in the novel session.
• For the session-field reading, transients outside the field are not removed one by one; the contiguous active bins around the field stand in for it.
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

We then measured spatial correlation of place fields before and after DCZ/CNO administration and found a reduction at all locations across the track, similar to the effects in RE low (Fig. 4d–g and Supplementary Fig. 17A–C ). This decrease was not observed in the saline control

Results, Bilateral VTA inhibition largely replicates lowering reward expectation, paragraph 3; Figure 4d to g; Supplementary Fig. 17a to c

What the paper reports

• per position, UR against other conditions: two tailed KS-test, p < 0.01 (n = R/RE low 928, Before/After Saline 1139, Before/After DCZ 1629, Before/After CNO 1498 place cells)

Under L1 (engaged RE low laps), mean correlations are R/RE low 0.351, saline 0.793, DCZ 0.657 and CNO 0.653; DCZ and CNO are below saline (KS p = 1.9e-16, 3.3e-12). In every position group saline is above R/RE low (KS p 4.3e-21 to 2.7e-28) and DCZ and CNO lie below saline; the same under every reading. The place cell counts are 610, 676, 842 and 641, outside the tolerances of 928, 1139, 1629 and 1498 (the dandiset has 4 of the 6 mice).

How the test differed from the paper’s method

• The dandiset holds 4 of the 6 DREADD mice, all 4 with saline (paper 5).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The DREADD files carry a constant offset of about 17000 counts, removed as the file minimum before dF/F. Assumption.
• Behaviour of sub-014 from BeforeCNO on is realigned by 6000 frames (found in the notebook).
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

This anticipatory licking behavior continued for a few laps in UR (5 ± 1 lap) and decayed exponentially (Fig. 1e ) except for some animals (4/12) that randomly licked in later laps

Methods, Licking behavior; Results, Lowering reward expectation changes spatial encoding, paragraph 1; Supplementary Fig. 1

RE high laps per mouse average 6.60 +/- 2.62 under L1, 7.60 under L2 and 8.60 under L3, outside 5 +/- 1 under every reading; per mouse 2 to 12 laps (L1). Counting the rewarded first UR lap of NR6 and NR14 as R gives 6.40 to 8.40. Random licking after the lick stop, read as a reward-zone lick in an RE low lap (reading a), is found in 4 of 10 mice, the paper's count, under every lick stop reading; reading (b), any lick on the track, gives 7, and reading (c) 9. The criterion needs both the mean and the count, and the mean fails. The index of the first UR lap with no zone lick, a reading the criterion does not list, averages 4.80 +/- 1.99, within 5 +/- 1; it is shown, not judged, and asked of a reader. The reward zone (first lick before the reward) starts at 184.5 to 196.9 cm in 8 mice, 167.7 cm in NR14 and 129.0 cm in CFC17. With the code's reward zone (0 cm in every mouse, as licks after the teleport fill the first bin) the means are 6.8, 7.8 and 8.8.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.

On average (mean ± 95% CI), this ratio was 1.01 ± 0.03 in the Dark, 1.3 ± 0.02 in the Rewarded condition and 1.22 ± 0.02 in the Unrewarded condition

Methods, Engagement with VR, approach behavior; Supplementary Fig. 6a

Only R and UR could be tested; the dark sessions are not in the dandiset. The ratio of mean binned velocity at 100 to 150 cm over 175 to 200 cm, taken over the traversal (lap start to the first frame at the track end, main reading), is 2.214 +/- 0.092 in R (352 laps, 1 undefined) and 1.745 +/- 0.072 in UR (297 laps), over mice 2.224 and 1.735, against 1.3 and 1.22. Over the whole lap, which takes in the 1.5 s pause at 200 cm, it is 5.17 and 4.44, with 74 R and 76 UR laps undefined (zero mean speed in the end bins) and ratios up to 150.7. The code's ratio (velocity normalised to its peak) over the traversal gives 1.259 and 1.199, the closest but outside the tolerance of 0.02. On a synthetic lap with a true ratio of 1.600 the helper returns 1.601 over the traversal and 1.895 over the whole lap. In every reading R is above UR. No measured value is given for the dark.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm. Assumption. The 30 Hz clock agrees with the 1.5 s pause after the reward (46 to 49 frames from reward onset to teleport in every rewarded condition), and lap time does not depend on the scale.
• The dark sessions of the claim are not in the dandiset.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.

In UR we observed changes in velocity throughout the track, but approach behavior remained intact on most laps in UR ( n = 12, 170/244 laps, 70%)

Results, Changes in spatial encoding associated with diminished reward expectation are not due to disengagement, paragraph 1; Methods, Engagement with VR, approach behavior

Engaged laps by the Methods ratio over the traversal (main reading), 10 mice. Under L1, of 231 RE low laps 106 (46%) are engaged and 125 disengaged, none undefined, with 66 RE high laps; L2 gives 100, 121 and 76, L3 92, 119 and 86. Only the L3 RE high count is within tolerance (90 +/- 9); the engaged and disengaged counts are outside 170 +/- 17 and 74 +/- 7.4 under every reading. The mean ratios of engaged and disengaged laps are 2.007 and 1.362 against 1.3 and 1.02. The code's ratio over the traversal (variant, not judged) gives 154 engaged and 77 disengaged laps under L1 (67%), both within tolerance, with ratios 1.284 and 1.029, but 66 RE high laps. The whole-lap reading of the first version (163 and 68) counted 62 laps with an undefined ratio as disengaged; with them left out it has 6 disengaged laps.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm. Assumption. The 30 Hz clock agrees with the 1.5 s pause after the reward (46 to 49 frames from reward onset to teleport in every rewarded condition), and lap time does not depend on the scale.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.

The lap-wise cluster probability traces revealed dips in probability when transitioning from RE high to RE low and not between R and RE high

Results, Lowering reward expectation induces place cell remapping, paragraph 1; Figure 2a

The 2 qualifying mice (NR6, NR23). In NR6 the R-cluster probability is 0.87 in R, 0.10 in RE high and 0.05 in RE low (L1): it falls at the first UR lap, not at the lick stop. In NR23, with 2 to 4 RE high laps, it is 0.82, 0.49 and 0.23 (L1), RE high already more than 0.2 below R. The rule holds in 0 of 2 mice under every lick stop reading; the criterion asks for 2 of 3. With the code-ratio split (L1c, 4 mice, not judged) it holds in 2 of 4 (NR21 0.84, 0.82, 0.15; CFC4 0.89, 0.72, 0.14). RR probabilities are 0.02 to 0.33. Keeping only the first 24 of NR6's 43 R laps, as the paper's code does (task_framestokeep), NR6 still fails under every reading (R 0.96, RE high 0.08 to 0.10, RE low 0.09 to 0.12); NR6's field values with fields found on those 24 laps were not computed.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• k-means from scikit-learn with one random start per run and 1000 seeded runs.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

We found 27.9% of place fields found in R (605 PFs in R, n = 3) remained stable throughout R-RE low -RR (169/605; Fig. 2e blue throughout). In RE low , R place fields either remained stable (222/605, 36.7%; Fig. 2e blue, middle column), disappeared (249/605, 41.2%; Fig. 2e black, middle column), or remapped (134/605, 22.1%; Fig. 2e cyan, middle column)

Results, Lowering reward expectation induces place cell remapping, paragraph 3; Figure 2e

Under L1 and F1, in the 2 qualifying mice - R 245 fields, RE low engaged 151, RR 254; stable through 11.0%, stable in RE low 19.6%, lost 75.9%, remapped 4.5%, new RE low 60.3%; stable in RR 40.4%, remapped 9.0%; lost fields reappearing 34.9%; new RE low fields stable in RR 41.8% and lost 53.8%; new RR fields 35.8%. The lost and new percentages do not depend on the fate definition. Under F1 to F3 and L1 to L3, 0 or 1 of the 11 percentages is within tolerance; under L1c (4 mice) 0, with 77.5% lost.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• New RR fields are counted as new to both R and RE low (main) and new to RE low (shown); the paper does not say.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

Plotting place fields defined in each condition, we found a total number of place fields that tiled the track to be similar in R, RE low (using only Engaged laps), and RR

Results, Place fields show diminished reliability, paragraph 1; Figure 3a; Figure 3 legend

In the 2 qualifying mice, R has 245 fields, the engaged RE low laps 151 (L1; 149 under L2, 146 under L3) and RR 254, so RE low has 38% fewer fields than R; all three are outside the tolerances of 605, 663 and 674, which count 3 mice. Under L1c (4 mice) the counts are 747, 332 and 736; RR is within tolerance and RE low far below R.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

out/in place field firing ratio (Rewarded = 0.10 [0.07 0.13]; RE low : Engaged = 0.16 [0.12 0.20])

Results, Place fields show diminished reliability, paragraph 1; Figure 3b

What the paper reports

• R: 95% CI 0.07 to 0.13 (n = 3 mice)
• RE low engaged: 95% CI 0.12 to 0.2 (n = 3 mice)
• RR: 95% CI 0.08 to 0.15 (n = 3 mice)

Per-mouse means of the 2 qualifying mice under L1 - R 0.068, RE low 0.086, RR 0.075, against 0.10 +/- 0.03, 0.16 +/- 0.04 and 0.12 +/- 0.04; RE low is 0.087 and 0.092 under L2 and L3, and the pooled fields give the same picture, so no reading puts the three values in tolerance. RE low is above R (pooled, unpaired t p = 1.4e-03). Under L1c (4 mice) R 0.074 and RR 0.080 are within and RE low 0.105 is not. The first version passed under L2 with the RE low laps of the whole-lap engagement split.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

place field width (Rewarded = 53.34 [43.87 62.81]; RE low : Engaged = 60.56 [48.07 73.06]), across all locations (Fig. 3b ). There was also a small decrease observed in firing intensity (Rewarded = 0.36 [0.31 0.40]; RE low : Engaged = 0.31 [0.23 0.40])

Results, Place fields show diminished reliability, paragraph 1; Figure 3b

What the paper reports

• width, R: 95% CI 43.87 to 62.81 cm
• width, RE low engaged: 95% CI 48.07 to 73.06 cm
• width, RR: 95% CI 46.86 to 61.31 cm
• firing intensity, R: 95% CI 0.31 to 0.4
• firing intensity, RE low engaged: 95% CI 0.23 to 0.4
• firing intensity, RR: 95% CI 0.24 to 0.34

Per-mouse means of the 2 qualifying mice under L1 - width 70.1, 65.6 and 73.1 cm in R, RE low and RR against 53.34, 60.56 and 54.08 (only RE low within), and RE low is not wider than R (pooled p = 0.24); intensity 0.380, 0.311 and 0.393 against 0.36, 0.31 and 0.29 (RR outside), falling from R to RE low (pooled p = 8.7e-04). The candidate regions the field search cuts are 31 to 35 cm wide on average (L1 to L3), so the Methods width rule (field-only map above 0) roughly doubles them. Under L1c (4 mice) widths are 85.8, 82.1 and 79.0 cm.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• Width is the span of contiguous bins where the field-only mean map is above 0, as the Methods define it; with the transient rule assumed here the fields are 1.1 to 1.7 times as wide as the paper's (70.1, 65.6 and 73.1 cm against 53.34, 60.56 and 54.08 cm here; medians of 75 to 81 cm against 46 to 55 cm in supp11-all-mice-parameters). The candidate regions (31 to 35 cm) are about half as wide as the fields.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

Medians: Reliability - R = 0.48, RElow = 0.26, RR = 0.44. Out/In Field Firing - R = 0.10, RElow = 0.19, RR = 0.13. Place field width (cm) - R= 50.21, RElow = 54.96, RR = 45.71. Firing Intensity - R = 0.34, RElow = 0.26, RR = 0.28

Supplementary Fig. 11b

All 10 mice, all RE low laps after the lick stop (no engagement split). Reading (a), median of per-mouse means under L1 - reliability 0.565, 0.391, 0.506; out/in 0.075, 0.098, 0.082; width 76.95, 80.82, 77.10 cm; intensity 0.441, 0.320, 0.368 - puts 0 of 12 within 10%; reading (b), median of pooled fields, puts 3 of 12. Every direction holds (reliability and intensity lower and out/in and width higher in RE low than R) under both readings and all lick stop readings.

How the test differed from the paper’s method

• The dandiset holds 10 of the 12 extinction mice (5 of the 6 with RR and a novel session).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• The reward zone starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport and belong to the previous reward (Methods, Licking behavior, says 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls, as the first version did, moved the zone by 0.3 to 3.9 cm and changed no lick stop.
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

We similarly found an over-representation of place fields near the reward site in R (Fig. 3d–f left, I; Supplementary Fig. 11C–E ). Interestingly, this over-representation disappeared once reward expectation diminished

Results, Lowering reward expectation eradicates the over-representation of reward location, paragraph 1; Figure 3d to f; Methods, Reward over-representation

In the 2 qualifying mice under L1 the end minus middle excess is 6.8 and -4.2 points in R (not above 0 in NR23), -14.6 and 5.5 in RE low (mean -4.6; above R in NR23) and -5.7 and 9.5 in RR, so the per-mouse conditions fail under every lick stop reading. The Gaussian fits are within tolerance (R 106.4 +/- 58.6, RE low 108.4 +/- 54.7, RR 103.8 +/- 57.9 cm). Under L1c (4 mice) R is above 0 in 3 of 4 mice, RE low below R in 3 of 4 with a mean of -7.4, and RR above 0 in 3 of 4.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

Of the place cells in R that disappeared in RE low , 41.8% (104/249, n = 3) were at the end of the track (150–200 cm). Furthermore, the new place cells that appeared in RE low , were equally distributed on the track length (0–50 cm: 77/296 (26%), 50–100 cm: 83/296 (28%), 100–150 cm: 62/296 (21%), 150–200 cm: 74/296 (25%))

Results, Lowering reward expectation eradicates the over-representation of reward location, paragraph 1

Under L1 and F1, in the 2 qualifying mice - of 186 R fields lost in RE low, 50 (26.9%) are at 150 to 200 cm (paper 104 of 249, 41.8%); new RE low fields per 50 cm are 21, 16, 30 and 24 (23, 18, 33 and 26%; paper 77, 83, 62 and 74, 21 to 28%); 18 lost end fields reappear in RR (paper 61); new RR fields are 29, 18, 17 and 27 per 50 cm (paper 31, 24, 18 and 67). Most counts are outside tolerance under every reading, but two are inside - new RR fields at 0 to 50 cm (29, 31 and 29 under L1 to L3, against 31 +/- 3.1) and at 100 to 150 cm under L1 and L2 (17, against 18 +/- 1.8; 22 under L3). Under L1c (4 mice) 200 of 579 lost fields (34.5%) are at the end.

How the test differed from the paper’s method

• The mice of Figures 2 and 3 are the R-UR-RR mice with more than 12 engaged RE low laps (Methods, Engagement with VR, approach behavior): NR6 and NR23 under every lick stop reading with the main engagement split, 2 mice against the paper's 3. With the code's ratio (reading L1c, shown, not judged) they are NR6, NR21, NR23 and CFC4. The files do not say which mice the paper used.
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.

DCZ and CNO administration caused a decrease in lap running speed in R (Supplementary Fig. 16 , Mean speed (m/s) [95% CI]: Before Saline 42.74 [41.12 44.36], After Saline 42.75 [41.34 44.17], Before DCZ 41.99 [40.26 43.73], After DCZ 25.96 [24.30 27.62], Before CNO 43.18 [41.26 45.10], After CNO 25.71 [24.03 27.39]) but approach behavior (demonstrating engagement) and anticipatory licking remained

Results, Bilateral VTA inhibition largely replicates lowering reward expectation, paragraph 3; Supplementary Fig. 16

What the paper reports

• before saline: 95% CI 41.12 to 44.36
• after saline: 95% CI 41.34 to 44.17
• before DCZ: 95% CI 40.26 to 43.73
• after DCZ: 95% CI 24.3 to 27.62
• before CNO: 95% CI 41.26 to 45.1
• after CNO: 95% CI 24.03 to 27.39

Lap speed (200 cm over the time from 0 to 200 cm), laps pooled over the 4 mice - before and after saline 25.94 and 26.15, DCZ 27.24 and 13.54, CNO 27.60 and 14.56 cm/s; per-mouse means and the mean running speed are also all outside tolerance (0 of 6 under every reading). DCZ and CNO lower lap speed (-13.70 and -13.05 cm/s, unpaired t p = 7.9e-38 and 2.0e-29) and saline does not (+0.22, p = 0.78), and nearly all laps after DCZ and CNO keep a pre-reward lick in the zone. The criterion needs the six means within tolerance, so the claim is not reproduced although the pattern holds. The 30 Hz clock agrees with the 1.5 s reward pause and lap speed does not depend on the volts-to-cm scale, so the slower speeds are in the data or in an undefined lap speed measure.

How the test differed from the paper’s method

• The dandiset holds 4 of the 6 DREADD mice, all 4 with saline (paper 5).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm. Assumption. The 30 Hz clock agrees with the 1.5 s pause after the reward (46 to 49 frames from reward onset to teleport in every rewarded condition), and lap time does not depend on the scale.
• Lap speed is read in cm/s, as the criterion reads the paper's m/s.
• Behaviour of sub-014 from BeforeCNO on is realigned by 6000 frames (found in the notebook).

We found a small reduction in spatial correlation on the slow velocity laps compared to the fast laps (Supplementary Fig. 18B, C ), but this was much smaller than the effects induced by DCZ and CNO

Results, Bilateral VTA inhibition largely replicates lowering reward expectation, paragraphs 3 and 4; Supplementary Fig. 18

What the paper reports

• reliability, before saline fast laps: 95% CI 0.46 to 0.57
• reliability, after saline slow laps: 95% CI 0.32 to 0.58
• out/in ratio, before saline fast laps: 95% CI 0.04 to 0.13
• out/in ratio, after saline slow laps: 95% CI 0.04 to 0.14

Percentiles within each mouse, reading (a) - fast before-saline laps 30.63 cm/s (paper 47.30), slow after-saline laps 21.17 (paper 35.03); reliability per mouse 0.340 and 0.366 (paper 0.51 and 0.45), out/in 0.066 and 0.076 (within 0.09 +/- 0.05). Reading (b) 30.63 and 29.41 cm/s, reliability 0.340 and 0.433. Pooled percentiles (thresholds 28.0 and 24.4 cm/s) give 32.56 and 18.36 cm/s (a) with reliability 0.351 and 0.400 per mouse, and 0.438 for the fast laps over pooled fields. Reliability of the fast laps is outside tolerance under all four splits. The drop in spatial correlation from the saline pair (0.793) to the split is 0.217 (a), 0.141 (b), 0.520 (pooled a) and -0.019 (pooled b), against half the DCZ drop, 0.068; only pooled (b), where 55 of the 62 fast laps come from 2 mice, is below it. From the same fast-lap fields against all after-saline laps the drop to the slow laps is 0.086 (a) and 0.010 (b).

How the test differed from the paper’s method

• The dandiset holds 4 of the 6 DREADD mice, all 4 with saline (paper 5).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm. Assumption. The 30 Hz clock agrees with the 1.5 s pause after the reward (46 to 49 frames from reward onset to teleport in every rewarded condition), and lap time does not depend on the scale.
• Percentiles are taken within each mouse and over the laps of the 4 mice pooled; both are run, as the criterion asks.
• The drop is measured from the correlation of the whole saline pair; the paper does not say what the split is compared with. The split's fields are defined on the fast laps only, so the drop from the same fields against all after-saline laps is also shown.
• Behaviour of sub-014 from BeforeCNO on is realigned by 6000 frames (found in the notebook).

We found a similar decrease in place field reliability and increase in out-of-field firing in DCZ and CNO as in RE low (Fig. 4h, j , Supplementary Fig. 17D , Mean [95%CI]: Reliability: Rewarded = 0.41 [0.31 0.52], RE low = 0.29 [0.17 0.40], Before DCZ = 0.48 [0.38 0.58], After DCZ = 0.41 [0.31 0.51], Before CNO = 0.42 [0.31 0.52], After CNO = 0.35 [0.24 0.47]. Out/In Field Firing: Rewarded = 0.10 [0.06 0.13], RE low = 0.17 [0.06 0.29], Before DCZ = 0.10 [0.05 0.14], After DCZ = 0.12 [0.08 0.16], Before CNO = 0.10 [0.05 0.15], After CNO = 0.12 [0.07 0.17]). This was not the case in the saline controls (Fig. 4i , Mean [95%CI]: Reliability: Before Saline = 0.45 [0.34 0.55], After Saline = 0.46 [0.39 0.54], Out/In Field Firing: Before Saline = 0.09 [0.04 0.14], After Saline = 0.09 [0.04 0.14])

Results, Bilateral VTA inhibition largely replicates lowering reward expectation, paragraph 4; Figure 4h to j; Supplementary Fig. 17d

What the paper reports

• reliability, R: 95% CI 0.31 to 0.52
• reliability, RE low: 95% CI 0.17 to 0.4
• reliability, before DCZ: 95% CI 0.38 to 0.58
• reliability, after DCZ: 95% CI 0.31 to 0.51
• reliability, before CNO: 95% CI 0.31 to 0.52
• reliability, after CNO: 95% CI 0.24 to 0.47
• reliability, before saline: 95% CI 0.34 to 0.55
• reliability, after saline: 95% CI 0.39 to 0.54
• out/in ratio, R: 95% CI 0.06 to 0.13
• out/in ratio, RE low: 95% CI 0.06 to 0.29
• out/in ratio, before DCZ: 95% CI 0.05 to 0.14
• out/in ratio, after DCZ: 95% CI 0.08 to 0.16
• out/in ratio, before CNO: 95% CI 0.05 to 0.15
• out/in ratio, after CNO: 95% CI 0.07 to 0.17
• out/in ratio, before saline: 95% CI 0.04 to 0.14
• out/in ratio, after saline: 95% CI 0.04 to 0.14

Per-mouse means under L1 (engaged RE low laps) - all 16 values within tolerance (also pooled). Out/in rises after DCZ and CNO; reliability falls after DCZ (0.452 to 0.438) but not after CNO (0.414 to 0.417) over per-mouse means, while over pooled fields both fall. Saline is not without change: out/in falls after saline (0.091 to 0.081, paired t over 4 mice p = 0.013; over cumulative histogram bins p = 8.0e-08) and reliability differs over the cumulative histogram bins (p = 4.5e-06). The claim that saline does not change these measures fails under both pairings, and per mouse CNO does not lower reliability.

How the test differed from the paper’s method

• The dandiset holds 4 of the 6 DREADD mice, all 4 with saline (paper 5).
• dF/F, significant transients and the place field search follow ref. 75 (Dombeck et al. 2010), which the Methods cite without the details (Methods, Image processing and ROI selection; Defining place fields) - 8th percentile baseline in a 15 s window, transients from 2 to 0.5 robust SD lasting 0.5 s or more, candidate fields above 25% of the way from baseline to peak, 100 circular-shift shuffles. Assumption, as the files hold raw fluorescence only and the paper does not give these values.
• Engaged laps are defined by the Methods approach ratio taken over the traversal of the track, from the lap start to the first frame at the track end, which leaves out the 1.5 s pause after the reward (Methods, Engagement with VR, 'instantaneous velocity was calculated at each point along the 2 m track'); laps whose ratio is undefined are neither engaged nor disengaged. The Methods do not say where the lap ends, so this is an assumption. It gives 46% engaged RE low laps under L1 and 2 of the 5 R-UR-RR mice with more than 12 engaged laps (paper 70%, 3 of 6); the code's ratio over the traversal, a variant, gives 67% and 4 mice. The whole-lap reading of the first version took in the pause and counted laps with an undefined ratio as disengaged; it is shown as a variant only.
• The DREADD offset is removed as for fig4f-dreadd-spatial-correlation. Assumption.
• The reward zone, which sets the lick stop and so the RE high and RE low laps, starts at the mean position of the first lick before the reward in each R lap, ignoring licks before 20 cm, which follow the teleport (Methods, Licking behavior, 'the average track position at which the animal started pre-licking in R'). The 20 cm is an assumption. Counting the first lick whenever it falls moves the zone by 0.3 to 3.9 cm and changes no lick stop under L1, L2 or L3 (printed in lick-persistence-ur).
• Position is stored in volts with no scale; the top of the range, where the reward is delivered at the track end (Methods, Behavior and virtual reality), is read as 200 cm, and the 5 cm position bins, field positions and widths rest on that scale. Assumption.
• Behaviour of sub-014 from BeforeCNO on is realigned by 6000 frames (found in the notebook).