Li, H., Namburi, P., Olson, J.M. et al. Neurotensin orchestrates valence assignment in the amygdala. Nature 608, 586-592 (2022). (DANDI:000689)

Claims

From the paper Li, H., Namburi, P., Olson, J.M. et al. Neurotensin orchestrates valence assignment in the amygdala. Nature 608, 586-592 (2022).

Reproducing Li et al. (2022) from dandiset 000689 7

Notebook reproduce_li_2022.ipynb

Partially reproduced2

Partially reproducedResults, NT inputs from the PVT amplify BLA representation of learned valences, paragraph 1; Figure 4bverifier disagreed at first
685 BLA neurons were recorded from 30 mice during the three-cue discrimination task, with sucrose, shock and neutral CSs at a 2 to 1 to 1 ratio.
We recorded a total of 685 BLA neurons from 30 mice while they were performing a discrimination task
Results, NT inputs from the PVT amplify BLA representation of learned valences, paragraph 1; Figure 4b
The dandiset holds 31 recording sessions from 31 mice, within the tolerance of 30. It holds 1902 units; 1739 have a baseline rate (1 s before each CS) of at least 0.3 Hz, and 1884 a whole-session rate of at least 0.3 Hz. Neither unit count is within 68.5 of 685, so the mouse count holds and the neuron count does not. The units are not sorted single units (finding), with a median of 64 per mouse against the paper's 23. The 2 to 1 to 1 ratio of the statement holds in 25 of 31 sessions; H09 to H14 have no neutral CS.
How the test differed from the paper’s method
- The paper analyses sorted single units (Offline Sorter); the files' units are used as they are.
Partially reproducedResults, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 3; Figure 5d-e and 5g-h; Methods, Unsupervised trial-based classification of behaviorsrests on an assumption
Unsupervised trial-by-trial clustering of pose-derived behavior gives seven clusters for sucrose trials and five for shock trials, which separate passive and active strategies.
Following watershed segmentation, we obtained seven clusters for sucrose trials and five clusters for shock trials
Results, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 3; Figure 5d-e and 5g-h; Methods, Unsupervised trial-based classification of behaviors
The pose has no timestamps, but a linear map of the trial clock onto the pose clock (offset and rate factor) fitted to the port entries holds in 21 of 29 sessions (best extra lag within 1 s in each third; 20 of the 21 factors between 0.9874 and 0.9885; median head-to-port distance at port entries 38 pixels against 105 over all frames). On the 1319 sucrose and 688 shock trials of those sessions (300 frames after each CS, 6264 tsfresh features, 5931 and 5927 kept), no reading gives both counts; each of the four meets one and not the other. With features z-scored and the installed scikit-learn defaults, the reading adopted, the kernel density has 7 peaks for sucrose and 3 for shock for all 5 seeds; with the 2020 defaults 7, 7, 8, 7, 8 and 4, 3, 4, 3, 3. With raw features it has 6 for sucrose and 5 for shock for all 5 seeds (installed defaults), and 7, 5, 7, 5, 6 and 5, 5, 5, 6, 7 (2020 defaults). So the sucrose count holds and the shock count does not under the z-scored features, and the reverse under the raw ones, and the outcome is partially_reproduced under every reading. The raw-feature counts are not stable. In the raw features the 3 largest features carry 95% (sucrose) and 98% (shock) of the variance and the 10 largest all of it; with the installed defaults an exact PCA in place of the randomized one gives 8 and 5, and multiplying each feature by 1 + 1e-9 noise gives 7 and 7. With z-scored features the same changes give 7 and 4, so the sucrose count stays at 7 and the shock count stays below 5. The run reviewed in round 2, with the same code on another runner, printed 7 and 5 (raw, installed defaults) and 9 and 4 (z-scored, installed defaults), which shows the same sensitivity. Without H08, H10 and H24, whose clock is checked least well (head-to-port distance under the map 29.0, 27.7 and 38.4 px against 35.1, 34.8 and 46.8 px over all frames), the adopted reading gives 5 and 4. The statement's "separate passive and active strategies" is not tested.
How the test differed from the paper’s method
- The pose is put on the trial clock by a linear map fitted to the port entries (nb.fit_pose_clock), since the files give only a start time and a nominal 15 frames per second; this is an assumption. H11, H14, H17, H23, H25, H29, H39 and H40, where the map does not hold, and H28 and H32, whose recorded session has no pose, are left out.
- tsfresh 0.16.0 does not install with Python 3.13; tsfresh 0.21.2 with its comprehensive default feature set is used (6264 features per trial, the paper's "~6000").
- Features that are not finite or are constant are dropped. Whether the features were scaled before PCA is not said (Methods, Unsupervised trial-based classification of behaviors); both are run and give the same outcome. z-scoring is the reading adopted for the measured counts, since on the raw features 10 features carry all the variance of the 30 PCs and the counts change with the PCA solver and with changes of 1e-9 in the features.
- The criterion's "defaults of scikit-learn" are read as those of the installed release (1.9, PCA initialisation, automatic learning rate), as the criterion says; the defaults of the releases contemporary with tsfresh 0.16.0 (2020; random initialisation, learning rate 200) are run as a reading and give the same outcome. Under either set the tsfresh release is 0.21.2, not 0.16.0. The kernel density is seaborn's default (scipy gaussian_kde, Scott's bandwidth, a 200 by 200 grid reaching 3 bandwidths beyond the data); each local maximum is a watershed marker, so the count of maxima is the count of regions. Counting only maxima above 5% of the highest gives the same counts.

Not reproduced3

Not reproducedResults, Genetically-encoded fluorescent NT sensor, paragraph 1; Figure 3kverifier disagreed at firstrests on an assumption
The BLA NT sensor response to sucrose delivery increases after reward learning.
After learning, we observed an increased NT sensor response to sucrose
Results, Genetically-encoded fluorescent NT sensor, paragraph 1; Figure 3k
What the paper reports
- US window, late against early: paired t-test, two-tailed, t(4) = 8.709, p = 0.001, 95% CI 2.744 to 5.538 (n = 5 mice)
- CS window, late against early: paired t-test, two-tailed, t(4) = 1.788, p = 0.1483, 95% CI -1.523 to 7.031 (n = 5 mice)
Five mice, early against late acquisition sessions as the files label them. Under the per-trial z-score (the reading adopted) the US-window response (2 to 7 s) is larger late than early under all four readings of the criterion. With all trials it is significant, AUC +3.53 (95% CI 1.82 to 5.24), t(4) = 5.74, p = 0.0046, 5 of 5 mice higher, inside the paper's CI of 2.744 to 5.538 around 4.14 (mean z +0.71, same test). With the first 16 trials it is not, AUC +2.94 (95% CI -0.77 to 6.66), t(4) = 2.20, p = 0.093, 4 of 5 mice higher. AUC and mean give the same t and p, so the two passing readings are one result. The criterion asks for p < 0.05 with late higher under every reading; it holds in 2 of the 4, so the criterion is not met, and only the direction holds in all four. The significant result also rests on the regression fitted on the trial-averaged baseline, whose correlation is 0.48 to 0.75 in four of the ten sucrose sessions (M-0 early and late, M-1 late, M-3 early). With the regression fitted on the pooled baseline bins of every trial (correlation at least 0.938) no reading reaches p < 0.05 (AUC +1.45 to +2.53, p >= 0.44), nor under the session-pooled z-score (AUC -2.92 to +0.16, p >= 0.16). The CS window does not change under any reading (p >= 0.25), as in the paper (P = 0.1483). The first run read the 470 nm channel of Mouse6's late sucrose session from the constant third frame type (see deviations); with that corrected the all-trials readings pass. Round 2 kept the outcome as partially_reproduced and proposed not_reproduced; round 3 agreed, since the criterion has no partial condition.
How the test differed from the paper’s method
- The files mark one kind of cue, 2 s long, and no US. All cues are taken as CS-US trials with the US 2 s after CS onset (Methods, Head-fixed behavior, "The US ... was delivered 2 s after the onset of the CS"). The Methods say about half the CSs were unpaired; in the late sucrose sessions 49 to 58 of about 60 cues are followed by licking within 2 to 5 s, so most cues are paired, but unpaired cues, if any, cannot be removed. This is an assumption.
- The median filter of 200 ms is applied at the native 20 Hz before averaging into 5 Hz bins, since at 5 Hz it would do nothing (Methods, Fiber photometry, Data analysis, "down-sampled to 5Hz and filtered ... using a median filter with a span of 200 ms").
- Whether the residuals are z-scored against each trial's baseline or the session's pooled baseline is not said (Methods, "Residuals from the regression were z-transformed using data a baseline window of -5 s to 0 s relative to the start of the CS"). The per-trial baseline is adopted as an assumption (each trial's own -5 to 0 s window is the plainer reading); the session-pooled one is shown and gives no p < 0.05.
- The 470 nm trace is regressed on the 415 nm trace over the trial-averaged -5 to 0 s window, as the Methods say ("Regression coefficients were computed using data averaged across trials in a session"). The fitted slopes run from 0.38 to 2.33 (correlation 0.25 to 0.99); a fit on the pooled baseline bins of every trial is shown as a variant.
- The single photometry series interleaves three frame types at 60 Hz where the Methods give 40 Hz. Each channel is the phase of the three-frame order that the 415 or 470 nm LED labels mark before the first irregular frame interval. In Mouse6's late sucrose session the labels move by one frame after a 2.3 ms interval while the data keep their order (phase means 0.200, 0.019 and 0.057 before and after it), so the labels there would give the constant third frame type as the 470 nm channel.
Not reproducedResults, Genetically-encoded fluorescent NT sensor, paragraph 1; Figure 3lrests on an assumption
The BLA NT sensor response to airpuff decreases after punishment learning.
and a decreased NT sensor response to airpuff
Results, Genetically-encoded fluorescent NT sensor, paragraph 1; Figure 3l
What the paper reports
- US window, late against early: paired t-test, two-tailed, t(4) = 3.996, p = 0.0162, 95% CI -6.083 to -0.9583 (n = 5 mice)
- CS window, late against early: paired t-test, two-tailed, t(4) = 2.137, p = 0.0994, 95% CI -2.585 to 0.357 (n = 5 mice)
Under the per-trial z-score the US-window response to airpuff is lower late than early under all four readings (AUC -5.47, 95% CI -12.35 to 1.41, with the first 16 trials, 5 of 5 mice lower; -4.66 with all trials), near the paper's -3.657, but no reading reaches p < 0.05 (smallest p = 0.074, t(4) = -2.40). Under the session-pooled z-score p >= 0.096, and with the regression fitted on the pooled baseline bins p >= 0.23. The CS window does not change significantly under any reading (p >= 0.08), as in the paper (P = 0.0994). In Mouse1's early airpuff session (29 cues) the last three cues come 4.7 and 7.4 s apart, one a 4.7 s bout; the first-16 readings do not include them.
How the test differed from the paper’s method
- The files mark no airpuff, and the cue events last 2 s where the Methods give a 4 s CS. All cues are taken as CS-US trials with the airpuff 2 s after CS onset (Methods, Head-fixed behavior, "presented 2 s after the onset of the CS"). The Methods say about half the CSs were unpaired; with about 60 cues per airpuff session where the Methods give about 30 pairings, up to half the trials analysed may be unpaired, and the files cannot tell them apart. This is an assumption.
- Same processing choices as fig3k-nt-sensor-sucrose-increase (median filter at 20 Hz, per-trial z-score adopted with the session one shown, regression on the trial-averaged baseline with the pooled fit shown, channels from the frame order).
Not reproducedResults, Genetically-encoded fluorescent NT sensor, paragraph 1rests on an assumptionquestion to the reader
Early in acquisition, the BLA NT sensor signal transiently decreases after both sucrose and airpuff.
Initially, we observed transient decreases in NT sensor responses to both sucrose and airpuff during early acquisition trials
Results, Genetically-encoded fluorescent NT sensor, paragraph 1
In the early sessions the mean over 5 mice of the US-window z-score is below zero for sucrose (-0.62 with the first 16 trials, per-trial z-score; 3 of 5 mice below zero) but above zero for airpuff (+0.11; 4 of 5 mice are below zero, but M-4 is at +2.15). No reading has both means below zero (0 of 8). Under the session-pooled z-score with the first 16 trials the signs reverse (sucrose +0.34, airpuff -0.40). No mean differs from zero (p >= 0.23). Read as a brief dip, the lowest point of the mouse-averaged trace in 2 to 7 s is below zero for both outcomes under the per-trial z-score (-1.09 for sucrose and -0.63 for airpuff with the first 16 trials, -0.82 and -0.36 with all trials). Whether the window mean tests a "transient" decrease is asked of a reader (questions/000689.yaml); the outcome is judged on the criterion as written.
How the test differed from the paper’s method
- Same trial assumption and processing choices as fig3k-nt-sensor-sucrose-increase and fig3l-nt-sensor-airpuff-decrease.

Blocked2

BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4c· Metadata insufficientverifier disagreed at first
BLA-NAc and BLA-CeA neurons are identified by short-latency responses to light at the end of each session.
At the end of each session, we optogenetically photoidentified BLA-NAc and BLA-CeA neurons based on previously reported photoresponse latencies
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4c
No file gives the projection of each mouse, so the BLA-NAc and BLA-CeA counts cannot be split, but both are within 10% of 39 and 32 only if their total is within 63.9 to 78.1. With the Methods' rule and the fake-session test, block 1 taken as the fake session as the Methods order them, the total is 146 under reading (a) (all 5 and 10 ms pulses and the onsets of the 100 ms light events) and 133 under reading (b) (isolated 1 Hz pulses). With block 2 as the fake one it is 256 and 233; letting H31 to H42, where one block precedes the task, take either order gives at least 104 and 102; with a baseline rate of at least 0.3 Hz the lowest is 91. Every reading is above 78.1, so the counts cannot both match whatever the projections. But the fake-session test does not work as the Methods describe it. Without it, block 1 evokes a short-latency response in 1353 of 1902 units and block 2 in 1243 (reading a; 1474 and 1374 under reading b), and 88% of the units responsive to block 2 also respond to block 1. A session with the patch cord disconnected should evoke none, so either neither block is that fake session or the responses are light artefacts on units that are not sorted single units (finding). The 91 to 256 tagged units are then units that cross the threshold in one block and not the other, and their count does not test the claim. The files label neither the fake and real blocks nor the projection of each mouse, so the claim is blocked. Round 2 kept the outcome as not_reproduced and proposed blocked (metadata_insufficient); round 3 agreed.
How the test differed from the paper’s method
- Which light block is the fake session is not labelled. Block 1 is taken as the fake one (Methods, In vivo electrophysiology, "a phototagging session was conducted after a 'fake' phototagging session"); the reverse, and either order in H31 to H42, are run as readings and agree.
- The files hold light events of 5, 10 and 100 ms and none of 1 s; the 100 ms events stand in for the long light in reading (a). Reading (b) does not use them and gives the same outcome.
- The count is compared as a total, since the projection labels are missing.
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4d· Methods underspecified
Unsupervised hierarchical clustering of BLA responses to the sucrose, neutral and shock CSs gives seven functional clusters.
we clustered BLA neurons into seven functional groups based on their responses across all task events using unsupervised hierarchical clustering
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4d
Run on the 1399 units of 24 mice with baseline of at least 0.3 Hz in sessions with all three CSs (163 units dropped for rate, 340 in the six sessions without neutral CSs, the whole of H17 for rate since its spikes stop at 249 s). With Ward linkage cut at 35% of the maximum distance, reading (a) gives 9, 12, 11, 12, 11, 10, 8 and 9 clusters for smoothing windows of 20 to 420 ms, and 6, 7, 6, 6 and 8 for 620 ms to 2 s; reading (b) gives 8 to 14 and never 7. Seven clusters come only under reading (a) with an 820 ms window. The paper's "smoothing factor of 0.85" is a MATLAB smoothdata setting whose window the paper does not give and that cannot be derived without MATLAB's heuristic, and the count moves with the window, so no reading has a defensible basis over another. The count also moves with the cut (14 at 25%, 13 at 30%, 10 at 35%, 7 at 40%, 5 at 50% for a 220 ms window). Without H22, H27, H34 and H39, whose CS alignment is not shown, seven clusters come under (a) at 420 ms and under (b) at 1020 ms and 2 s. Some clusters are single sessions (at 220 ms one cluster is 93% H27), so a count here would not show functional types. The question is for the paper's authors (finding). The units are not the paper's sorted single units, which alone could change the count.
How the test differed from the paper’s method
- The smoothing window is run from 20 ms to 2 s, as the paper gives only MATLAB's smoothing factor (Methods, Functional hierarchical clustering, "a Gaussian-weighted moving average with a smoothing factor of 0.85"); each CS trace is smoothed separately.
- The six sessions without neutral CS trials (H09 to H14) are left out, since the clustered trace concatenates the three CSs (Methods, "Data from each experiment protocol (Sucrose CS, Neutral CS, Shock CS) were concatenated").

Not tested in a notebook 32

19 data missing from the dataset, 13 metadata insufficient

Blocked32

BlockedResults, NT inputs to the BLA co-release glutamate, paragraph 1; Figure 1a-c· Data missing from the dataset
NT neurons projecting to the BLA are found in the medial geniculate nucleus (MGN), the ventral hippocampus (vHPC) and the paraventricular nucleus of the thalamus (PVT).
We observed that the medial geniculate nucleus (MGN), ventral subiculum and ventral CA1 of the hippocampus (vHPC), and the paraventricular nucleus of the thalamus (PVT) had NT neurons (tdTomato expressing) projecting to the BLA
Results, NT inputs to the BLA co-release glutamate, paragraph 1; Figure 1a-c
The retrograde tracing in NT::Cre;Ai14 mice is not in the dandiset, which holds BLA recordings, pose, confocal images of PVT slices from the recorded mice and NT sensor photometry. The paper names no other archive for these data.
BlockedExtended Data Figure 1c and 1f· Data missing from the dataset
Ntsr1 mRNA is more often colocalized with BLA-CeM projectors than with BLA-NAc projectors, while Ntsr2 mRNA does not differ between them.
Ntsr1 mRNA was predominantly expressed in BLA-CeM neurons compared to BLA-NAc neurons
Extended Data Figure 1c and 1f
What the paper reports
- Ntsr1, colocalization as a fraction of tracer-labelled cells: unpaired t-test, t(5) = 7.901, p = 5 × 10⁻⁴, 95% CI 0.1737 to 0.3412
- Ntsr1, colocalization as a fraction of Ntsr1-positive cells: unpaired t-test, t(5) = 6.383, p = 0.0014, 95% CI 0.2613 to 0.6137
- Ntsr2, BLA-CeA against BLA-NAc: unpaired t-test, two-tailed, t(5) = 0.3729, p = 0.7245, 95% CI -0.1202 to 0.08974
The RNAscope images of BLA projectors are not in the dandiset.
BlockedResults, NT inputs to the BLA co-release glutamate, paragraph 1; Extended Data Figure 2d-e· Data missing from the dataset
Light-evoked EPSCs in BLA neurons from MGN:NT, vHPC:NT and PVT:NT axon terminals persist in TTX and 4-AP (monosynaptic) and are reduced by AMPA and NMDA receptor blockers, while IPSCs are abolished by TTX and 4-AP.
the EPSCs evoked in BLA neurons by stimulating ChR2-expressing axon terminals from all three NT populations were monosynaptic and can be blocked by antagonizing the AMPA and NMDA glutamate receptors ( Figure E2d ), whereas the IPSCs were not
Results, NT inputs to the BLA co-release glutamate, paragraph 1; Extended Data Figure 2d-e
What the paper reports
- MGN, TTX+4-AP against AP5+NBQX: one-way ANOVA with Holm-Sidak's multiple comparisons, t(27) = 3.19, p = 0.0072
- PVT, TTX+4-AP against AP5+NBQX: one-way ANOVA with Holm-Sidak's multiple comparisons, t(39) = 4.585, p < 10⁻⁴
- vHPC, TTX+4-AP against AP5+NBQX (t = 28.03 as printed does not fit P = 0.0188 or the effect size 224 with SE 79.91, which give t near 2.8): one-way ANOVA with Holm-Sidak's multiple comparisons, t(26) = 28.03, p = 0.0188
The slice patch-clamp recordings are not in the dandiset (no intracellular data in any file).
BlockedResults, Global Nt gene knockout in the PVT promotes reward learning, paragraph 1; Extended Data Figure 3a and 3c· Data missing from the dataset
The two Nt guide RNAs with Cas9 produce indel mutations in about 90% of sequenced reads in vivo (90.8% for guide 1 and 80.3% for guide 2 above control tissue).
We verified that our guide RNAs (guide RNA 1 and guide RNA 2) had high editing efficiency (~90 %) in vivo
Results, Global Nt gene knockout in the PVT promotes reward learning, paragraph 1; Extended Data Figure 3a and 3c
What the paper reports
- guide 1 against control: unpaired t-test, two-tailed, t(8) = 85.99, p < 10⁻⁴, 95% CI 88.38 to 93.25
- guide 2 against control: unpaired t-test, two-tailed, t(7) = 21.20, p < 10⁻⁴, 95% CI 71.32 to 89.23
The sequencing data of the guide RNA editing are not in the dandiset.
BlockedExtended Data Figure 5f· Data missing from the dataset
The PVT-BLA Nt CRISPR-cKO reduces Nt mRNA in guide-positive PVT cells without changing Vglut2 mRNA.
Mice receiving Nt guides injections (CRISPR group) showed significantly reduced Nt mRNA levels in the guide-positive cells compared to mice receiving control guide injections, while Vglut2 mRNA levels were unaffected
Extended Data Figure 5f
What the paper reports
- Nt mRNA: unpaired t-test, t(28) = 2.788, p = 0.009, 95% CI -2.18 to -0.3335
- Vglut2 mRNA: unpaired t-test, t(12) = 0.724, p = 0.483, 95% CI -0.3528 to 0.7039
No in situ quantification is in the dandiset. The ecephys files of 27 mice (all but H01, H02, H07 and H08) hold stacks of 8 to 42 confocal images (COFAImages) whose descriptions name an Olympus FLUOVIEW file of the PVT (data check). The images are named only by depth, and no file says which mouse is CRISPR or control, what each image shows (channel or probe), or which cells are guide-positive, so they cannot stand in for the paper's RNAscope cell counts.
BlockedResults, Global Nt gene knockout in the PVT promotes reward learning, paragraph 2; Figure 1g· Data missing from the dataset
Global Nt CRISPR-cKO in the PVT, but not in the MGN or vHPC, increases port entry probability in the last tone-sucrose session.
We found that the CRISPR-Cas9 mediated conditional knockout (CRISPR-cKO) of Nt gene in the PVT, but not the MGN nor the vHPC, enhanced tone-sucrose association
Results, Global Nt gene knockout in the PVT promotes reward learning, paragraph 2; Figure 1g
What the paper reports
- PVT, CRISPR against control: unpaired t-test, two-tailed, t(26) = 2.611, p = 0.0148, 95% CI 4.363 to 36.67 (n = 15 CRISPR and 13 control mice)
- MGN, CRISPR against control: unpaired t-test, two-tailed, t(33) = 0.1019, p = 0.9195, 95% CI -16.1 to 17.8 (n = 18 CRISPR and 17 control mice)
- vHPC, CRISPR against control: unpaired t-test, two-tailed, t(18) = 0.8666, p = 0.3976, 95% CI -9.054 to 21.77 (n = 11 CRISPR and 9 control mice)
The global Nt cKO behavior cohorts (MGN, vHPC, PVT) are not in the dandiset.
BlockedFigure 1h· Data missing from the dataset
Global Nt CRISPR-cKO in the MGN, vHPC or PVT does not change freezing in the second session of tone-shock conditioning.
CRISPR-cKO in none of the regions affected tone-shock acquisition in second session of tone-shock conditioning
Figure 1h
What the paper reports
- MGN: unpaired t-test, two-tailed, t(24) = 0.8346, p = 0.4122, 95% CI -0.2184 to 0.09263 (n = 14 CRISPR and 12 control mice)
- vHPC: unpaired t-test, two-tailed, t(15) = 0.2820, p = 0.7818, 95% CI -0.1962 to 0.2561 (n = 8 CRISPR and 9 control mice)
- PVT (t(14) printed for 8 and 9 mice): unpaired t-test, two-tailed, t(14) = 0.5740, p = 0.5751, 95% CI -0.1585 to 0.2743 (n = 8 CRISPR and 9 control mice)
The global Nt cKO tone-shock sessions are not in the dandiset.
BlockedResults, PVT-BLA:NT neurons are a functionally-distinct PVT population, paragraph 1; Figure 2c· Metadata insufficient
PVT-BLA projection-specific Nt CRISPR-cKO reduces port entry probability to the sucrose-predictive CS relative to scrambled-guide controls.
The CRISPR group showed significantly reduced port entry probability in response to the sucrose-predictive CS
Results, PVT-BLA:NT neurons are a functionally-distinct PVT population, paragraph 1; Figure 2c
What the paper reports
- unpaired t-test, two-tailed, t(34) = 2.988, p = 0.0052, 95% CI -1.073 to -0.2042 (n = 17 CRISPR and 19 control mice)
The criterion's session reading (a), the last tone-sucrose day, is not in the dandiset (data missing); reading (b), the sucrose trials of the last discrimination day, is in the 31 recording sessions (port entries are in the trial tables), but no file says which mouse is CRISPR and which control, so no reading can be run.
BlockedFigure 2d· Data missing from the dataset
PVT-BLA Nt CRISPR-cKO increases freezing to the shock-predictive CS in the tone-shock test session.
The CRISPR group also showed significantly higher proportion of time spent freezing during the duration of shock-predictive CS in the tone-shock testing session
Figure 2d
What the paper reports
- unpaired t-test, two-tailed, t(26) = 2.167, p = 0.0396, 95% CI 0.01324 to 0.5002 (n = 14 CRISPR and 14 control mice)
The tone-shock test session of the PVT-BLA cKO cohort is not in the dandiset, which holds only discrimination-task sessions; the group labels are also missing (see fig2c).
BlockedFigure 2g· Data missing from the dataset
Photostimulating PVT:NT terminals in the BLA during the tones increases port entry probability to the sucrose-predictive CS relative to mCherry controls.
The ChR2 group showed significantly enhanced port entry probability in response to sucrose-predictive CS
Figure 2g
What the paper reports
- unpaired t-test, two-tailed, t(13) = 2.706, p = 0.018, 95% CI 0.2156 to 1.921 (n = 7 ChR2 and 8 mCherry mice)
The PVT-BLA:NT photostimulation behavior cohort is not in the dandiset.
BlockedFigure 2h (its statistics are printed under "2f" in Methods, Additional statistical information)· Data missing from the dataset
Photostimulating PVT:NT terminals in the BLA during conditioning reduces freezing to the shock-predictive CS in the tone-shock test session.
The ChR2 group showed significantly reduced freezing in the tone-shock testing session during the presentation of shock-predictive CS compared to the control group
Figure 2h (its statistics are printed under "2f" in Methods, Additional statistical information)
What the paper reports
- unpaired t-test, two-tailed, t(13) = 2.474, p = 0.0279, 95% CI -0.546 to -0.03698 (n = 7 ChR2 and 8 mCherry mice)
The PVT-BLA:NT photostimulation behavior cohort is not in the dandiset.
BlockedFigure 3b and 3c· Data missing from the dataset
Head-fixed mice acquire the tasks, with more anticipatory licks to the sucrose tone and more anticipatory eye closure to the airpuff tone in the last session than in the first.
averaged anticipatory licks to the sucrose tone for the first 16 trials of first and last tone-sucrose sessions
Figure 3b and 3c
What the paper reports
- anticipatory licks, last against first session: paired t-test, two-tailed, t(12) = 6.497, p < 10⁻⁴, 95% CI 1.21 to 2.431 (n = 13 mice)
- anticipatory eye closure, last against first session: paired t-test, two-tailed, t(11) = 2.48, p = 0.0306, 95% CI 0.4852 to 8.136 (n = 12 mice)
The anticipatory licks and the eye videos of the axon-terminal photometry cohort (13 and 12 mice) are not in the dandiset. The NT sensor files hold lick events for 5 other mice but no eye video.
BlockedResults, PVT-BLA:NT neurons are a functionally-distinct PVT population, paragraph 3; Figure 3e· Data missing from the dataset
PVT-BLA:NT axon-terminal calcium responses to sucrose delivery increase after tone-sucrose learning, while the CS response does not change significantly.
Calcium responses of the PVT-BLA:NT axon terminals to sucrose were significantly enhanced by reward conditioning
Results, PVT-BLA:NT neurons are a functionally-distinct PVT population, paragraph 3; Figure 3e
What the paper reports
- US window, late against early: paired t-test, two-tailed, t(12) = 2.935, p = 0.0125, 95% CI 3.635 to 24.6 (n = 13 mice)
- CS window, late against early: paired t-test, two-tailed, t(12) = 1.474, p = 0.1663, 95% CI -1.86 to 9.633 (n = 13 mice)
The PVT-BLA:NT axon-terminal GCaMP photometry is not in the dandiset; its only photometry is the GRAB-NTS1.0 sensor (fluorophore "neurotensin-fluorescent sensor", 415 and 470 nm, 5 mice).
BlockedFigure 3f· Data missing from the dataset
PVT-BLA:NT axon-terminal calcium responses to airpuff decrease after tone-airpuff learning.
Responses to airpuff were significantly suppressed by punishment conditioning
Figure 3f
What the paper reports
- US window, late against early: paired t-test, two-tailed, t(11) = 2.896, p = 0.0145, 95% CI -12.46 to -1.7 (n = 12 mice)
- CS window, late against early: paired t-test, two-tailed, t(11) = 1.472, p = 0.169, 95% CI -5.28 to 1.047 (n = 12 mice)
As fig3e-terminal-sucrose-increase.
BlockedResults, PVT-BLA:NT neurons are a functionally-distinct PVT population, paragraph 3; Extended Data Figure 6b-c· Data missing from the dataset
Somatic calcium responses of MGN, vHPC and PVT NT neurons from CS to US onset do not change between early and late acquisition of either task.
while somatic calcium responses of the NT three sources including PVT:NT had no learning-induced changes
Results, PVT-BLA:NT neurons are a functionally-distinct PVT population, paragraph 3; Extended Data Figure 6b-c
What the paper reports
- PVT, tone-sucrose: paired t-test, two-tailed, t(7) = 1.76, p = 0.1218, 95% CI -10.65 to 1.56
- PVT, tone-airpuff: paired t-test, two-tailed, t(7) = 0.5676, p = 0.588, 95% CI -4.608 to 7.519
- MGN, tone-sucrose: paired t-test, two-tailed, t(8) = 0.3214, p = 0.7561, 95% CI -6.528 to 4.931
- vHPC, tone-sucrose: paired t-test, two-tailed, t(7) = 1.615, p = 0.1504, 95% CI -3.558 to 0.6703
- MGN, tone-airpuff: paired t-test, two-tailed, t(8) = 0.2102, p = 0.8388, 95% CI -7.774 to 9.333
- vHPC, tone-airpuff: paired t-test, two-tailed, t(7) = 0.1973, p = 0.8492, 95% CI -2.567 to 2.172
The somatic GCaMP photometry of MGN, vHPC and PVT NT neurons is not in the dandiset.
BlockedResults, Genetically-encoded fluorescent NT sensor, paragraph 1; Figure 3h· Data missing from the dataset
The GRAB-NTS1.0 sensor responds to NT and not to 13 other transmitters and neuropeptides in cultured neurons.
validated the sensor to show high sensitivity and specificity for NT
Results, Genetically-encoded fluorescent NT sensor, paragraph 1; Figure 3h
What the paper reports
- one-way ANOVA with Dunnett's multiple comparisons test, F(13, 28) = 1059, p < 10⁻⁴ (n = 3 coverslips per compound)
The sensor imaging in cultured neurons is not in the dandiset.
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Extended Data Figure 8d· Metadata insufficient
In the CRISPR group, untagged BLA neurons and BLA-CeA neurons have lower basal firing rates than in controls, while BLA-NAc neurons do not differ.
Both non-tagged BLA neurons and the BLA-CeA neurons showed significantly lower basal firing rates (in Hz) in the CRISPR group compared to the control group
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Extended Data Figure 8d
What the paper reports
- untagged neurons: unpaired t-test, two-tailed, t(611) = 3.132, p = 0.0018, 95% CI -2.108 to -0.4834
- BLA-CeA neurons: unpaired t-test, two-tailed, t(30) = 2.429, p = 0.021, 95% CI -6.961 to -0.6071
- BLA-NAc neurons: unpaired t-test, two-tailed, t(37) = 0.4665, p = 0.6436, 95% CI -6.226 to 3.896
Needs the CRISPR and control groups and the tagged BLA-NAc and BLA-CeA neurons; every file gives the genotype as "Wild type" and none names the group or the Cav2-Cre injection site. The tagged neurons of fig4c-photoidentification are not known either (that claim is blocked, since the light blocks and the projections are not labelled).
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Extended Data Figure 8f-g· Metadata insufficient
The CRISPR group has smaller proportions of neurons excited or inhibited by each of the three CSs, but the response magnitude of the responsive neurons does not differ from controls.
CRISPR mice showed overall reduced proportions, but not responsive amplitudes of excited and inhibited neurons to all three cues relative to the control group
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Extended Data Figure 8f-g
Compares CRISPR and control; no file gives the group (see ed8d-reduced-basal-firing).
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4e· Metadata insufficient
Within the task-relevant clusters, the CRISPR group has smaller responses to the sucrose and shock CSs than controls, while neutral-CS responses are largely unchanged.
the CRISPR-cKO decreased the proportions of CS-responsive neurons and attenuated the magnitude of neural responses to Sucrose and Shock CSs within each task-relevant cluster
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4e
What the paper reports
- cluster 1, sucrose CS: unpaired t-test, two-tailed, t(78) = 2.203, p = 0.0303
- cluster 2, shock CS: unpaired t-test, two-tailed, t(73) = 2.365, p = 0.02
- cluster 3, sucrose CS: unpaired t-test, two-tailed, t(50) = 2.071, p = 0.0435
- cluster 3, shock CS: unpaired t-test, two-tailed, t(50) = 2.076, p = 0.043
- cluster 4, sucrose CS: unpaired t-test, two-tailed, t(52) = 2.219, p = 0.031
- cluster 4, shock CS: unpaired t-test, two-tailed, t(52) = 2.225, p = 0.03
- cluster 6, shock CS: unpaired t-test, two-tailed, t(140) = 2.228, p = 0.0276
- cluster 7, neutral CS: unpaired t-test, two-tailed, t(167) = 2.924, p = 0.003
Compares CRISPR and control within clusters; no file gives the group. It also depends on fig4d-seven-clusters, blocked.
BlockedExtended Data Figure 8i· Metadata insufficient
In the CRISPR group, smaller proportions of BLA-NAc neurons fall in the clusters that preferentially encode positive valence (clusters 1 and 3), and smaller proportions of BLA-CeA neurons in those that preferentially encode negative valence (clusters 2 and 4), than in controls.
The proportion of BLA-NAc neurons in cluster 1 and 3 which preferentially encode positive valence was less in the CRISPR group compared to the control group
Extended Data Figure 8i
Needs the groups and the tagged projectors, which no file gives.
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4f-g; Extended Data Figure 8j· Metadata insufficient
In controls BLA-NAc neurons are excited and BLA-CeA neurons inhibited by the sucrose CS; the CRISPR-cKO removes the BLA-NAc excitation and the BLA-CeA inhibition.
BLA-NAc and BLA-CeA population in the CRISPR group showed significantly reduced excitation and inhibition to the sucrose CS compared to the control group, respectively
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4f-g; Extended Data Figure 8j
What the paper reports
- BLA-NAc, sucrose CS, CRISPR against control: unpaired t-test, two-tailed, t(37) = 2.06, p = 0.0465, 95% CI -4.596 to -0.0376
- BLA-CeA, sucrose CS, CRISPR against control: unpaired t-test, two-tailed, t(30) = 2.06, p = 0.0481, 95% CI 0.0009 to 2.042
- BLA-NAc, Extended Data Figure 8j: unpaired t-test, two-tailed, t(37) = 2.043, p = 0.0483, 95% CI -3.209 to -0.01291
- BLA-CeA, Extended Data Figure 8j: unpaired t-test, two-tailed, t(30) = 2.46, p = 0.0199, 95% CI 0.453 to 4.882
Needs the groups and the tagged projectors, which no file gives.
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4f-g· Metadata insufficient
In control mice BLA-NAc neurons are predominantly excited by the sucrose CS and BLA-CeA neurons by the shock CS.
we replicated our previous reports in control mice showing that BLA-NAc and BLA-CeA neurons are predominantly excited by Sucrose CS and Shock CS, respectively
Results, NT inputs from the PVT amplify BLA representation, paragraph 1; Figure 4f-g
Needs the control mice and the tagged projectors, which no file gives.
BlockedResults, NT inputs from the PVT amplify BLA representation, paragraph 2; Figure 5a· Metadata insufficient
Population trajectories of the CRISPR group are shorter for all three CSs, and the distance between sucrose and shock trajectories is smaller, than in controls.
trajectories in the CRISPR group showed significant decreases in total trajectory length and trajectory distance between sucrose and shock trials compared to the controls
Results, NT inputs from the PVT amplify BLA representation, paragraph 2; Figure 5a
What the paper reports
- trajectory length, sucrose CS: unpaired t-test, two-tailed, t(11) = 4.786, p < 10⁻⁴, 95% CI -9.963 to -4.284 (n = 12 control and 14 CRISPR leave-one-mouse-out iterations)
- trajectory length, neutral CS (printed t = 1.224 with P = 0.0002): unpaired t-test, two-tailed, t(11) = 1.224, p = 2 × 10⁻⁴, 95% CI -10.37 to -3.736 (n = 12 control and 14 CRISPR leave-one-mouse-out iterations)
- trajectory length, shock CS: unpaired t-test, two-tailed, t(11) = 14.99, p < 10⁻⁴, 95% CI -13.15 to -7.113 (n = 12 control and 14 CRISPR leave-one-mouse-out iterations)
Compares CRISPR and control trajectories; no file gives the group.
BlockedResults, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 2; Figure 5b· Metadata insufficient
A linear SVM decodes sucrose against shock trials from BLA population activity above label-shuffled chance in both control and CRISPR groups, with no significant difference between the groups.
We found that BLA neurons from both control and CRISPR groups could decode the trial type compared to label-shuffled trials, when the SVM classifier was trained and tested within-group
Results, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 2; Figure 5b
What the paper reports
- ANOVA over the four conditions: one-way ANOVA with Holm-Sidak's multiple comparisons test, F(3, 16) = 668.6, p < 10⁻⁴
- control against control shuffled: Holm-Sidak's multiple comparisons, t(16) = 33.45, p < 10⁻⁴
- CRISPR against CRISPR shuffled: Holm-Sidak's multiple comparisons, t(16) = 29.78, p < 10⁻⁴
- control against CRISPR (statistics section; the legend gives P = 0.2122): Holm-Sidak's multiple comparisons, t(16) = 1.592, p = 0.1309
Decodes within each group; no file gives the group. It also depends on fig5a-trajectories-reduced, blocked.
BlockedResults, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 2; Figure 5c· Metadata insufficient
A decoder trained on CRISPR data decodes control data, but a decoder trained on control data performs significantly worse on CRISPR data.
SVM decoder performance significantly dropped when the model was trained on the control data and tested on the CRISPR data, compared to trained on the CRISPR data and tested on the control data
Results, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 2; Figure 5c
What the paper reports
- ANOVA over the conditions: one-way ANOVA with Holm-Sidak's multiple comparisons test, F(3, 16) = 34196, p < 10⁻⁴
- trained on CRISPR tested on control, against trained on control tested on CRISPR (printed effect size -0.3114): Holm-Sidak's multiple comparisons, t(16) = 181.0, p < 10⁻⁴
- trained on CRISPR tested on control, against shuffled (printed effect size 0.1662): Holm-Sidak's multiple comparisons, t(16) = 96.57, p < 10⁻⁴
- trained on control tested on CRISPR, against shuffled (printed effect size 0.4769): Holm-Sidak's multiple comparisons, t(16) = 277.1, p < 10⁻⁴
Decodes across groups; no file gives the group.
BlockedResults, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 1; Extended Data Figure 8e· Metadata insufficient
CRISPR mice make more port entries at baseline and in the late sucrose-trial period, have a lower baseline-normalized port entry response to the sucrose CS, and freeze more to the shock CS relative to baseline.
showed significantly enhanced baseline port entries and reduced port entry specificity during sucrose trials, compared to the control group
Results, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 1; Extended Data Figure 8e
What the paper reports
- sucrose trials, raw port entry probability at baseline (P = 0.012 as printed is half the two-tailed P for t(33) = 2.37, about 0.024): unpaired t-test, two-tailed, t(33) = 2.37, p = 0.012, 95% CI 0.01974 to 0.2594
- sucrose trials, raw port entry probability 10 to 25 s after sucrose: unpaired t-test, two-tailed, t(33) = 3.797, p = 6 × 10⁻⁴, 95% CI 0.1099 to 0.3638
- sucrose trials, z-scored port entry probability: unpaired t-test, two-tailed, p = 0.0038, 95% CI -1.067 to -0.2225
- shock trials, z-scored freezing: unpaired t-test, two-tailed, t(33) = 2.495, p = 0.0178, 95% CI 0.229 to 2.254
- shock trials, raw freezing: unpaired t-test, two-tailed, t(33) = 0.1401, p = 0.8894, 95% CI -0.0771 to 0.06722
Compares CRISPR and control mice; no file gives the group. Its freezing part would also need the pose on the trial clock, which the files do not give; a fitted clock holds in 21 of 29 sessions (see fig5dh-behavior-clusters).
BlockedResults, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 3; Figure 5f and 5i· Metadata insufficient
Control mice show more active-strategy trials than CRISPR mice in both sucrose and shock trials.
We found that control mice preferred active behavioral strategies during both sucrose and shock trials ... CRISPR-cKO abolished this behavioral preference to active strategies
Results, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 3; Figure 5f and 5i
What the paper reports
- sucrose trials: unpaired t-test, two-tailed, t(28) = 2.25, p = 0.0325, 95% CI -0.3141 to -0.01474 (n = 15 mice per group)
- shock trials: unpaired t-test, two-tailed, t(28) = 2.366, p = 0.0251, 95% CI -0.6799 to -0.04895 (n = 15 mice per group)
Compares CRISPR and control mice and depends on fig5dh-behavior-clusters; both the group labels are missing. The pose is aligned by a fitted clock in 21 of 29 sessions only.
BlockedResults, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 3; Extended Data Figure 8o and 8s· Metadata insufficient
BLA responses to the CSs are larger in active trials, and the CRISPR-cKO reduces the responses of untagged BLA neurons in active trials for both sucrose and shock CSs.
Finally, we observed predominant BLA neural responses to CSs during the active behavioral trials, which were diminished by the
Results, NT inputs from the PVT to the BLA mediates passive-active behavioral strategies, paragraph 3; Extended Data Figure 8o and 8s
What the paper reports
- sucrose CS, untagged neurons, active trials, control against CRISPR: two-way ANOVA with Holm-Sidak's multiple comparisons test, t(1284) = 2.27, p = 0.046
- shock CS, untagged neurons, active trials, control against CRISPR: two-way ANOVA with Holm-Sidak's multiple comparisons test, t(780) = 2.245, p = 0.049
- sucrose CS, untagged neurons, main effect of state (passive or active): two-way ANOVA, F(1, 1284) = 6.238, p = 0.0126
- shock CS, untagged neurons, main effect of state (passive or active): two-way ANOVA, F(1, 780) = 0.628, p = 0.4282
Compares CRISPR and control neurons in active and passive trials; needs the group labels and the behavior clusters, neither available.
BlockedResults, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 9f-g· Data missing from the dataset
Bath-applied NT (10 nM) enhances EPSCs in BLA-NAc neurons and suppresses them in BLA-CeM neurons, and an NTSR1 antagonist abolishes the difference.
At a concentration of 10 nM, NT significantly enhanced EPSCs in BLA-NAc neurons and suppressed EPSCs in BLA-CeM neurons
Results, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 9f-g
What the paper reports
- NT alone, first population of the legend (W = -45, all ranks negative, so the suppressed population, read as BLA-CeM; the legend does not name it): Wilcoxon signed-rank test, W = -45, p = 0.0039 (n = 9 neurons)
- NT alone, second population of the legend (W = 52, mostly positive ranks, so the enhanced population, read as BLA-NAc): Wilcoxon signed-rank test, W = 52, p = 0.0425 (n = 12 neurons)
- with NTSR1 antagonist, first population of the legend: Wilcoxon signed-rank test, W = -26, p = 0.0781 (n = 8 neurons)
- with NTSR1 antagonist, second population of the legend: Wilcoxon signed-rank test, W = -13, p = 0.4961 (n = 9 neurons)
The slice recordings are not in the dandiset.
BlockedResults, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 9k-l· Data missing from the dataset
Pre-incubation with NT (10 nM) reduces STDP-induced LTP in BLA-NAc neurons but not in BLA-CeM neurons.
STDP-induced LTP in the presence of NT (10 nM) 10–50 minutes after induction was reduced in BLA-NAc neurons, but unchanged in BLA-CeM neurons compared with the ACSF control
Results, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 9k-l
What the paper reports
- BLA-NAc: unpaired t-test, t(31) = 2.177, p = 0.0367, 95% CI -0.649 to -0.022
- BLA-CeM: unpaired t-test, t(25) = 1.193, p = 0.2442, 95% CI -0.1305 to 0.4896
The slice recordings are not in the dandiset.
BlockedResults, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 10b-c· Data missing from the dataset
Intra-BLA NTSR1 antagonist before conditioning enhances tone-sucrose learning and does not change freezing in tone-shock conditioning.
NTSR1 antagonist delivered in the BLA 15 minutes before the conditioning session enhanced reward learning but had no effects on punishment learning
Results, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 10b-c
What the paper reports
- sucrose, session 2 normalized response: unpaired t-test, t(16) = 2.128, p = 0.049, 95% CI 0.0029 to 1.483
- freezing during the tone: unpaired t-test, t(19) = 1.061, p = 0.3021, 95% CI -7.577 to 23.15
The pharmacology behavior cohort is not in the dandiset.
BlockedResults, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 10k and 10o· Data missing from the dataset
Ntsr1 CRISPR-cKO in BLA-CeA neurons, but not BLA-NAc neurons, reduces port entry to the sucrose CS; global BLA Ntsr1 cKO before acquisition impairs tone-sucrose learning, while cKO after acquisition does not affect recall.
we found that CRISPR-cKO of Ntsr1 gene in the BLA suppressed reward learning
Results, The temporal dynamics of NT are critical for valence assignment, paragraph 1; Extended Data Figure 10k and 10o
What the paper reports
- BLA-CeA Ntsr1 cKO, port entry: unpaired t-test, two-tailed, t(16) = 3.577, p = 0.0025, 95% CI -0.7107 to -0.1818
- BLA-NAc Ntsr1 cKO, port entry: unpaired t-test, two-tailed, t(9) = 0.03604, p = 0.972, 95% CI -0.7141 to 0.7373
- global BLA cKO, ANOVA over control, pre-acquisition and post-acquisition groups: one-way ANOVA, F(2, 25) = 4.324, p = 0.024
- global BLA cKO, control against pre-acquisition cKO: one-way ANOVA, Holm-Sidak's multiple comparisons test, p = 0.019
- global BLA cKO, control against post-acquisition cKO: one-way ANOVA, Holm-Sidak's multiple comparisons test, p = 0.99
The Ntsr1 cKO behavior cohorts are not in the dandiset.

Findings about the data

What attempts found about the dataset itself, for its owner and the archive.

Description missingNo file says which recorded mouse is in the CRISPR group and which in the control group, nor whether its Cav2-Cre was injected in the NAc or the CeA; every subject's genotype is "Wild type". The paper's electrophysiology findings compare the groups and the tagged projectors, so 13 of its claims cannot be tested, and the phototagging count cannot be split by projection. Neither the units table (spike times and a unit name), the electrodes table (location "unknown"), any other text field of the files, nor the archive's subject metadata (age, sex, species, genotype) holds such a label.
File format errorThe units are not sorted single units, as the Methods describe (Offline Sorter). 21 of 31 sessions hold exactly 64 units on 32 channels, 1902 units in all against the paper's 685; in 1061 units more than 5% of inter-spike intervals are below 1 ms, and in each session 6 to 65% of spike samples recur in another unit. Both light blocks evoke a short-latency response in most of them (1353 and 1243 of 1902), so the light rule of the Methods with the fake-session test (91 to 256 tagged units, against 71 tagged neurons in the paper) cannot single out tagged neurons.
File format errorIn H17 the spike times of all 64 units fall within the first 249 s of a 12118 s recording, while its trial events cover the whole session, so its spikes cannot be aligned with its trials.
Description missingThe DeepLabCut pose has no timestamps, only a start time of 0 s and a nominal rate of 15 frames per second, and the mp4 videos give a constant 15 frames per second. Against the trial clock, port entries match the head's approach to the port at lags a median of 80 s and up to 192 s from zero that drift within 28 of 29 sessions. A linear map fitted to the port entries removes the drift in 21 sessions, with a rate factor of about 0.988 in 20 of them, as if the true frame rate were about 1.2% below 15; in 8 sessions no such map holds (lags of 2.8 to 19.2 s are left in some third). The videos of H07, H09 and H10 run 818 to 1528 s longer than their pose series, and that of H11 426 s longer; the others are within 10 s. Video timestamps, or the frame times of the original .asf files, would settle the alignment.
Description missing27 of the 31 ecephys files (all but H01, H02, H07 and H08) hold a stack of 8 to 42 confocal images (COFAImages) whose description names an Olympus FLUOVIEW file of the PVT. The images are named only by depth, with no channel, probe, stain or group label, so what they show cannot be read from the files.
Description missingEach session holds about 854 light events in two blocks of about 427, which fits the paper's "fake" phototagging session and the real one, but no label says which is which. In 12 sessions (H31 to H42) one block comes before the task, where the Methods put both after it. The two blocks evoke much the same short-latency responses (1353 and 1243 of 1902 units respond to the first and the second, and 88% of the units responsive to the second block also respond to the first), so neither behaves like a session with the patch cord disconnected. The files hold light events of 5, 10 and 100 ms and none of the 1 s light the Methods describe.
Metadata errorSix sessions (H09 to H14) hold 75 sucrose and 45 shock CSs and no neutral CS, while their session description says sucrose, shock and neutral CSs at 2 to 1 to 1.
OtherThe dandiset has recordings from 31 mice where the paper reports 30. For H28 and H32 the recorded session (Disc3, Disc5) is earlier than another discrimination day for which only pose is deposited (Disc4, Disc6), while the paper analyses only the last day.
Description missingThe photometry files store one RoiResponseSeries at 60 Hz that interleaves the 415 nm frames, the 470 nm frames and a third, undescribed frame type; the channels can be told apart only from the LED event labels, and in one session (Mouse6, late sucrose) the labels move by one frame after a 2.3 ms frame interval while the data keep their order, so after it they mark the constant third frame type as 470 nm. In Mouse1's early airpuff session the last three cues come 4.7 and 7.4 s apart, one lasting 4.7 s. The cue events last 2 s where the Methods give a 4 s CS, and no event marks the US or says whether a cue was paired. In two airpuff sessions the lick label is on for most of the session.
Data missingOnly the in vivo recordings of Figures 4 and 5 and the NT sensor photometry of Figure 3j-l are deposited. The behavior cohorts of Figures 1 and 2 and Extended Data Figure 10, the axon-terminal and somatic photometry with its head-fixed behavior (Figure 3b-f, Extended Data Figure 6), the anatomy and in situ data, the slice physiology and the sensor validation are not, so 19 of the paper's 39 findings in the claims file cannot be tested.
Methods not availableQuestion for the paper's authors. The functional clustering smooths each PSTH with MATLAB smoothdata and a smoothing factor of 0.85 (Methods, Functional hierarchical clustering). What window did that produce, and was the smoothing applied to each CS trace or to the concatenated trace? On the dandiset's units the number of clusters under the paper's cut-off ranges from 6 to 12 over windows of 20 ms to 2 s, not monotonically (9 at 20 ms, 12 at 60 ms, 6 at 620 ms, 8 at 2 s), with seven only at 820 ms.
Methods not availableQuestion for the paper's authors. The NT sensor residuals were "z-transformed using data a baseline window of -5 s to 0 s" (Methods, Fiber photometry, Data analysis). Was each trial z-scored against its own baseline, or against the session's baseline pooled over trials? The sucrose effect of Figure 3k reaches p = 0.0046 with each trial's baseline and all trials, and p >= 0.16 with the pooled baseline.
Methods not availableQuestion for the paper's authors. For the behavior clustering (Methods, Unsupervised trial-based classification of behaviors), were the tsfresh features scaled before PCA, and which scikit-learn release and t-SNE settings (perplexity, initialisation, learning rate) were used? On the dandiset's pose the sucrose and shock cluster counts range from 5 to 8 and 3 to 7 across these choices, the PCA solver and the seeds, and no reading gives 7 and 5 together. On unscaled features 10 features carry all the variance, and the counts change with the PCA solver and with changes of 1e-9 in the features, and between two runs of the same code on different runners.

Questions

Where the author or a verifier believed the claims file misread the paper, they asked, citing the paper, and a reader who had not seen the data answered from the paper alone, amending the claims file where it was wrong. The author then ran the claim as answered. Attempts before protocol pilot-11 also asked here for method details the claims file did not give.

openasked 2026-10-31about fig3-nt-sensor-early-dip
The criterion judges the "transient decreases" by the mean z-scored signal over the whole US window (2 to 7 s from CS onset). Does a mean over 5 s test a transient decrease, or should the criterion look at the lowest point of the trace in that window (or a shorter window after the US)? A brief dip can be averaged away by a later rise.
Passage. Results, Genetically-encoded fluorescent NT sensor, paragraph 1, "Initially, we observed transient decreases in NT sensor responses to both sucrose and airpuff during early acquisition trials". The paper reports no test or window for this finding; the 2 to 7 s window is that of the Figure 3 legend for the late against early comparisons.
Why it matters. In this attempt the mouse-averaged early airpuff trace falls to -0.63 (z, per-trial baseline, first 16 trials) inside 2 to 7 s, and the early sucrose trace to -1.09, while the mean over the window is +0.11 for airpuff. Under the window mean the claim is not reproduced; under the lowest point it would be, for both outcomes. The verifier of round 1 reported this as a possible misreading.
No reader has answered it yet.

Notebooks

Every registered entry that uses this dataset. An entry can hold several notebooks, and can use more than one dataset.

Reproduction Neurotensin and valence assignment in the amygdala (Li, Namburi, Olson et al. 2022)

Tests the findings of Li, Namburi, Olson et al. (2022) that dandiset 000689 can hold. It holds 31 sessions of BLA recordings during the three-cue discrimination task, with pose, and the NT sensor photometry of 5 mice during reward and punishment learning. The NT sensor learning effects, the neuron and tagged-unit counts, the functional clustering and the behavior clustering are tested; the group comparisons are blocked because no file says which mice are CRISPR and which control, and the other experiments of the paper are not in the dandiset. All files

Reproducing Li et al. (2022) from dandiset 000689
reproduce_li_2022.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-31
Agreed with 38 of 39 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: Round 1. The first executed notebook with the conclusion cells and the summary removed, the helper module and the claims file, with the paper. The verifier queried the archive asset metadata and tested photometry_trials on synthetic data. Report in transcripts/000689/verifier-1-report.md.
- fig4-recording-count: the verifier’s outcome was “Partially reproduced”. Mice, 31, inside 30 +/- 3. Neurons, 1902 before the 0.3 Hz exclusion and 1739 after (cell 29), both far outside 685 +/- 68.5. Six sessions (H09 to H14) have no neutral CS, so the 2 to 1 to 1 ratio holds in only 25 of 31 sessions (cell 5).
Round 2
2026-10-31
Agreed with 37 of 39 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: Round 2. The full executed notebook after the round-1 revision, with the conclusions and the summary, the helper module, the claims file and the attempt record, with the paper. Report in transcripts/000689/verifier-2-report.md.
- fig3k-nt-sensor-sucrose-increase: the verifier’s outcome was “Not reproduced”. Should be not_reproduced. The criterion says "Reproduced when the US comparison has p < 0.05 with late higher under every reading". It has one part, and only 2 of its 4 readings pass (cell 25). Those 2 are really one result, because AUC and mean give the same t and p. With the first 16 trials p = 0.093. The direction matches in all four readings, and the record can say so. The per-trial z-score is a reasonable assumption, but p = 0.0046 also depends on the trial-averaged regression, which has r = 0.25 to 0.75 in all four M-0 sessions and in M-1 late sucrose (cell 23); the pooled fit (r >= 0.94) gives p = 0.64 with all trials. The record reports this, but the conclusion depends on this one fit choice. The README sentence that the response rises after learning with p = 0.005 when all trials are used holds only under the per-trial z-score with the trial-averaged fit (session z-score p = 0.16, pooled fit p = 0.64). Cell 27, "partially reproduced, holds in 2 of its 4 readings", contradicts the criterion's "under every reading".
- fig4c-photoidentification: the verifier’s outcome was “Blocked”. Block 1, taken as the fake session, gets a short-latency response from 1353 of 1902 units, and 88% of block-2 responders also respond to block 1 (cell 42). So it does not behave like a session with the patch cord disconnected, or the responses are light artefacts on unsorted channels. The 91 to 256 tagged units are then units that cross the threshold in one block but not the other, on units the record itself calls unsorted. Cell 43 says a count matching the paper was not expected. This does not test the claim, so it should be blocked (format_error, or the fake/real labels as metadata_insufficient). "Block 1 is the fake one" is an assumption not flagged assumed; all the orders were run, so this does not matter for the result.
Round 3
2026-10-31Agreed with 2 of 2 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: Round 3. The two changes of outcome proposed after round 2, on fig3k-nt-sensor-sucrose-increase and fig4c-photoidentification, and the notebook outputs behind them. Report in transcripts/000689/verifier-3-report.md.

Method differences

How the tests differed from the paper’s method or data, and the claims each applies to.

The files mark one kind of cue, 2 s long, and no US. All cues are taken as CS-US trials with the US 2 s after CS onset (Methods, Head-fixed behavior, "The US ... was delivered 2 s after the onset of the CS"). The Methods say about half the CSs were unpaired; in the late sucrose sessions 49 to 58 of about 60 cues are followed by licking within 2 to 5 s, so most cues are paired, but unpaired cues, if any, cannot be removed. This is an assumption.fig3k-nt-sensor-sucrose-increase
The median filter of 200 ms is applied at the native 20 Hz before averaging into 5 Hz bins, since at 5 Hz it would do nothing (Methods, Fiber photometry, Data analysis, "down-sampled to 5Hz and filtered ... using a median filter with a span of 200 ms").fig3k-nt-sensor-sucrose-increase
Whether the residuals are z-scored against each trial's baseline or the session's pooled baseline is not said (Methods, "Residuals from the regression were z-transformed using data a baseline window of -5 s to 0 s relative to the start of the CS"). The per-trial baseline is adopted as an assumption (each trial's own -5 to 0 s window is the plainer reading); the session-pooled one is shown and gives no p < 0.05.fig3k-nt-sensor-sucrose-increase
The 470 nm trace is regressed on the 415 nm trace over the trial-averaged -5 to 0 s window, as the Methods say ("Regression coefficients were computed using data averaged across trials in a session"). The fitted slopes run from 0.38 to 2.33 (correlation 0.25 to 0.99); a fit on the pooled baseline bins of every trial is shown as a variant.fig3k-nt-sensor-sucrose-increase
The single photometry series interleaves three frame types at 60 Hz where the Methods give 40 Hz. Each channel is the phase of the three-frame order that the 415 or 470 nm LED labels mark before the first irregular frame interval. In Mouse6's late sucrose session the labels move by one frame after a 2.3 ms interval while the data keep their order (phase means 0.200, 0.019 and 0.057 before and after it), so the labels there would give the constant third frame type as the 470 nm channel.fig3k-nt-sensor-sucrose-increase
The files mark no airpuff, and the cue events last 2 s where the Methods give a 4 s CS. All cues are taken as CS-US trials with the airpuff 2 s after CS onset (Methods, Head-fixed behavior, "presented 2 s after the onset of the CS"). The Methods say about half the CSs were unpaired; with about 60 cues per airpuff session where the Methods give about 30 pairings, up to half the trials analysed may be unpaired, and the files cannot tell them apart. This is an assumption.fig3l-nt-sensor-airpuff-decrease
Same processing choices as fig3k-nt-sensor-sucrose-increase (median filter at 20 Hz, per-trial z-score adopted with the session one shown, regression on the trial-averaged baseline with the pooled fit shown, channels from the frame order).fig3l-nt-sensor-airpuff-decrease
Same trial assumption and processing choices as fig3k-nt-sensor-sucrose-increase and fig3l-nt-sensor-airpuff-decrease.fig3-nt-sensor-early-dip
The paper analyses sorted single units (Offline Sorter); the files' units are used as they are.fig4-recording-count
Which light block is the fake session is not labelled. Block 1 is taken as the fake one (Methods, In vivo electrophysiology, "a phototagging session was conducted after a 'fake' phototagging session"); the reverse, and either order in H31 to H42, are run as readings and agree.fig4c-photoidentification
The files hold light events of 5, 10 and 100 ms and none of 1 s; the 100 ms events stand in for the long light in reading (a). Reading (b) does not use them and gives the same outcome.fig4c-photoidentification
The count is compared as a total, since the projection labels are missing.fig4c-photoidentification
The smoothing window is run from 20 ms to 2 s, as the paper gives only MATLAB's smoothing factor (Methods, Functional hierarchical clustering, "a Gaussian-weighted moving average with a smoothing factor of 0.85"); each CS trace is smoothed separately.fig4d-seven-clusters
The six sessions without neutral CS trials (H09 to H14) are left out, since the clustered trace concatenates the three CSs (Methods, "Data from each experiment protocol (Sucrose CS, Neutral CS, Shock CS) were concatenated").fig4d-seven-clusters
The pose is put on the trial clock by a linear map fitted to the port entries (nb.fit_pose_clock), since the files give only a start time and a nominal 15 frames per second; this is an assumption. H11, H14, H17, H23, H25, H29, H39 and H40, where the map does not hold, and H28 and H32, whose recorded session has no pose, are left out.fig5dh-behavior-clusters
tsfresh 0.16.0 does not install with Python 3.13; tsfresh 0.21.2 with its comprehensive default feature set is used (6264 features per trial, the paper's "~6000").fig5dh-behavior-clusters
Features that are not finite or are constant are dropped. Whether the features were scaled before PCA is not said (Methods, Unsupervised trial-based classification of behaviors); both are run and give the same outcome. z-scoring is the reading adopted for the measured counts, since on the raw features 10 features carry all the variance of the 30 PCs and the counts change with the PCA solver and with changes of 1e-9 in the features.fig5dh-behavior-clusters
The criterion's "defaults of scikit-learn" are read as those of the installed release (1.9, PCA initialisation, automatic learning rate), as the criterion says; the defaults of the releases contemporary with tsfresh 0.16.0 (2020; random initialisation, learning rate 200) are run as a reading and give the same outcome. Under either set the tsfresh release is 0.21.2, not 0.16.0. The kernel density is seaborn's default (scipy gaussian_kde, Scott's bandwidth, a 200 by 200 grid reaching 3 bandwidths beyond the data); each local maximum is a watershed marker, so the count of maxima is the count of regions. Counting only maxima above 5% of the highest gives the same counts.fig5dh-behavior-clusters

Provenance

Attempt record: 000689-li-2022-01
Paper: Li, H., Namburi, P., Olson, J.M. et al. Neurotensin orchestrates valence assignment in the amygdala. Nature 608, 586-592 (2022).
Date: 2026-10-31
Made by: An agent, run by CatalystNeuro
Dataset: DANDI:000689
Version: 0.240530.1923 (doi:10.48324/dandi.000689/0.240530.1923)
Run on: GitHub Actions
Harness: Claude Code
Model: claude-opus-5-5
Protocol: pilot-12

Transcripts of the agent runs

Each run’s prompt, messages and tool calls, with long output cut and the text of the paper left out, and its final report.

← DANDI:000688 DANDI:000691 →