We used juvenile male zebra finches (Taeniopygia guttata) 44–112 days post-hatch (dph) singing undirected song (n = 32 birds). Animals were not divided into experimental groups; thus, randomization and blinding were not necessary. No statistical methods were used to predetermine sample size. Birds were obtained from the Massachusetts Institute of Technology zebra finch breeding facility (Cambridge, Massachusetts). The care and experimental manipulation of the animals were carried out in accordance with guidelines of the National Institutes of Health and were reviewed and approved by the Massachusetts Institute of Technology Committee on Animal Care. All the juvenile birds were raised by their parents in individual breeding cages until 38 ± 5.2 dph (mean ± s.d.) when they were removed and were singly housed in custom-made sound isolation chambers (maintained on a 12:12 h day-night schedule). For a subset of the birds (birds 1, 2 and 4), additional tutoring was carried out after removal from the breeding cages to facilitate song imitation. This was done by playback of the tutor song through a speaker (20 bouts per day). Additional tutoring was done for 12 days for bird 1, 7 days for bird 2, and 18 days for bird 4. Bird identification key: bird 1, to3965; bird 2, to3779; bird 3, to3017; bird 4, to5640; bird 5, to3396; bird 6, to2309; bird 7, to3412; bird 8, to3567; bird 9, to2462; bird 10, to2331; bird 11, to2427; bird 12, to3352. To compare the activity of HVC projection neurons in juvenile birds with that of adult birds, we also included neurons recorded in adults (>120 dph, n = 3 birds) which included a reanalysis of previously published HVC recordings performed in adult male zebra finches singing directed song20. Songs were recorded with Sound Analysis Pro51 or a custom-written MATLAB software (A. Andalman), which was configured to ensure triggering of recordings on all quiet vocalizations of juvenile birds27. The vertical axis range for all spectrograms is 500–8,000 Hz. We classified each day of juvenile singing into one of four song stages: subsong stage, protosyllable stage, multi-syllable stage, and motif stage (Extended Data Fig. 1a). Subsong stage (48 ± 4 dph, median ± inter-quartile range, IQR) is defined as having a syllable duration distribution well-fit by an exponential distribution34, 35, with an upper limit for the Lilliefors goodness-of-fit statistic of 6. Following the subsong stage, birds enter the protosyllable stage (58 ± 10 dph, median ± IQR) characterized by the presence of syllables with consistent timing reflected in a peak in the distribution of syllable durations32, 33, 34, 35. The onset of the protosyllable stage was defined here as the first day in which the syllable duration distribution deviated from an exponential distribution (Lilliefors goodness-of-fit statistic greater than 6). Following the protosyllable stage, birds transition to the multi-syllable stage (62 ± 12 dph, median ± IQR) in which multiple distinct syllable types are visible in the song spectrogram and as multiple clusters in a scatter plot of syllable features52 (for example, Fig. 3a, b; 62 dph). The motif stage (73 ± 21 dph, median ± IQR) was defined by the production of a sequence of syllables in a relatively fixed order31. Finally, songs recorded in birds older than 120 dph were assigned as adult stage. A slightly older cutoff than the typical definition of adulthood in zebra finches (~90 dph)14 was used, because some of our birds in the 90–120 dph range continued to undergo some small developmental changes, as has been reported31. Syllable segmentation of the juvenile song was done based on the song power in a spectral band between 1 and 4 kHz, as described previously27, 34, 35. In a few cases, cutoff frequencies of the band-pass filters were adjusted to avoid the inclusion of high-frequency inspiratory sounds35, 53. Introductory notes were removed manually to avoid including HVC neurons that are rhythmically active during these elements54. Song bouts were defined as continuous sequences of syllables separated by gaps no longer than 300 ms35. Bout onset was defined as the onset of the first syllable in the bout, and bout offset was defined as the offset of the last syllable in the bout. For bird 3 (‘motif strategy’), it was difficult to segment syllables consistently using previous methods based on setting a threshold on the sound amplitude27, 34, 35. To overcome this limitation, we segmented syllables based on the phase of the rhythmicity in the song (‘phase segmentation’). The peak of the song rhythm, defined as the spectrum of the sound amplitude during singing38, exhibited a peak around 9 Hz (Extended Data Fig. 8c). To estimate the instantaneous phase of this rhythm, we first band-pass filtered the sound amplitude (Extended Data Fig. 8c, d; second-order IIR resonator filter with peak at 9 Hz and −3 dB half-bandwidth of 3 Hz; MATLAB command iirpeak). The band-pass filtered signal was then processed using the Hilbert transform (MATLAB command hilbert) to compute the instantaneous amplitude and phase (Extended Data Fig. 8d). Next, we set a threshold on this instantaneous amplitude to find the rhythmic part of the song. Finally, within this rhythmic part, song was segmented by detecting threshold crossings of the instantaneous phase (Extended Data Fig. 8d, bottom). Phase segments that contain no sounds or calls were manually removed. Similarly, phase segmentation (band-pass filter with peak at 10 Hz and half-bandwidth of 3 Hz) was used to segment the song during the protosyllable stage for bird 4 (Extended Data Fig. 9a, e, f). Note that this method is best suited for segmenting songs that have strong rhythmic modulation of song amplitude, but in which syllable boundaries are not strongly rhythmic. This appeared to be typical of birds employing the ‘motif strategy’32. Protosyllables were defined by their characteristic durations as has been described previously34, 35. In short, to identify the protosyllables, we first subtracted the best-fit exponential distribution (using 200–400 ms) from the syllable duration distribution, and fitted a Gaussian distribution to this residual. Protosyllables were defined as syllables having durations within two standard deviations from the mean of this Gaussian distribution. We labelled protosyllables using the Greek letter ‘α’ in all our birds for consistency. To label the emerging syllables in the juvenile song, we used the Greek letters β, γ, δ, and ε. In contrast, to label the syllables in the adult motif, we used the capital letters of the Latin alphabet A, B, C, etc. For birds in which the song learning trajectory was tracked developmentally, we labelled the syllables such that the correspondence between the juvenile syllables and adult syllables is straightforward: for example, α becomes A, β becomes B, γ becomes C, δ becomes D, and ε becomes E. Note that this labelling scheme leads to a slightly unconventional labelling of adult song in the sense that a motif can have letters in a reverse order (for example, CBA in Fig. 4f, g; Extended Data Fig. 6a), or a motif might not have a syllable A (for example, EDCB in Extended Data Fig. 7a). Syllable labelling was done manually by visual inspection of the song spectrogram; this was done blind with respect to the neural activity. The existence of multiple distinct syllable types were confirmed by calculating the syllable duration and acoustic features commonly used to analyse birdsong syllables51, 55, and visualizing the clusters of syllables in a two-dimensional space52 (Fig. 3b, Extended Data Figs 8b and 9d). In some cases, syllable order was used as an additional indicator of syllable identity (for example, Extended Data Fig. 7a, 70 dph; Extended Data Fig. 8a, 51 dph; Extended Data Fig. 9a, 59 dph). In bird 1, syllables β and γ were labelled manually by visual inspection of the song spectrogram (Fig. 3a). Since characterizing shared neurons and specific neurons depends on the reliable labelling of syllables, we took a conservative approach and only labelled syllables that were clearly identifiable and did not label the syllables that were ambiguous (fraction of syllables labelled as β or γ during 62–66 dph: 70 ± 5.5%, mean ± s.d.). We then estimated the error rate of our labelling procedure by plotting the labelled syllables (n = 200 syllables per type on each day) in a two-dimensional space of syllable duration and mean pitch goodness (Fig. 3b), and obtained a decision boundary using linear discriminant analysis. We used mismatch between manual labelling and feature-based labelling to estimate the error rate for syllables β and γ. The error rate during the first five days of syllable differentiation (62–66 dph), when the labelling was most difficult, was only 1.1% on average (range: 0.25–3.0%). For the second round of differentiation in bird 1, syllable order was used to assist in the labelling of syllables in early stages when syllables ‘B’ and ‘D’ were not easily distinguishable based on acoustic differences. Because these syllables underwent bout-onset differentiation, the first β after bout onset was labelled ‘D’; later renditions of β in the bout were labelled ‘B’ (Extended Data Fig. 7a). In bird 2, several emerging syllables could be easily distinguished based on syllable durations (Extended Data Fig. 6d). Specifically, syllables whose durations were 110–160 ms, and 180–250 ms were defined as α and β, respectively. Syllables that were 10–75 ms in duration were labelled γ if they were followed by a β, and labelled ε otherwise. Single-unit recordings of HVC projection neurons during singing were carried out using a motorized microdrive described previously56, 57. Single-units were confirmed by the existence of the refractory period in the inter-spike interval (ISI) distribution (Extended Data Fig. 1b). Neurons that were active only during distance calls and not during singing20 were excluded from the analysis. In addition, neurons recorded for less than 5 s of singing were excluded since the short recording duration did not allow us to reliably quantify the activity pattern of these neurons. Antidromic identification of HVC projection neurons was carried out with a bipolar stimulating electrode implanted in RA and Area X (single pulse of 200 μs every 1 s; current amplitude: 50–500 μA)19, 20, 57, 58, 59. A subset of antidromically identified projection neurons was further validated with collision testing19, 20, 57, 58, 59. A different subset of single units were identified as putative projection neurons based on sparse bursting, but could not be antidromically identified because they did not respond to antidromic stimulation or were lost before antidromic identification could be carried out (211 of 1,149 neurons). These neurons were included in the data set as unidentified HVC projection neurons (HVC ). HVC projection neurons exhibited bursts of action potentials during singing (Fig. 1a–c). The bursting nature of these neurons was evident in the inter-spike interval (ISI) distribution during singing, which exhibited two peaks with an inter-peak minimum near 30 ms (Extended Data Fig. 1b). We defined a ‘burst’ as a continuous group of spikes separated by intervals of 30 ms or less. Thus, by definition, bursts are separated from other spikes by intervals greater than 30 ms. Note that single spikes separated by more than 30 ms from both the preceding spike and the following spikes were also counted as a burst. Burst time was defined as the centre of mass of all the spikes within the burst. Burst width was defined as the interval between the first and the last spike in a burst (Extended Data Fig. 1c, top). Firing rate during burst was defined as the reciprocal of the mean inter-spike interval in a burst (Extended Data Fig. 1c, bottom). For the calculation of burst width and firing rate during bursts, bursts composed of a single spike were excluded. To analyse the temporal relation between neural activity and song syllables, we aligned the spike times to syllable onsets and constructed a rate histogram (1 ms bin, smoothed over 20 bins; range: ± 0.5 s from syllable onsets). The peak in this rate histogram was found between 50 ms before syllable onset and 200 ms after syllable onset. To test the significance of this peak, surrogate histograms were created by adding different random time shifts to the spike times on each trial60. Random time shifts were drawn from a uniform distribution over ± 0.5 s. The peak of this surrogate histogram was recorded, and this shuffling procedure was repeated 1,000 times; P values were obtained by analysing the frequency with which the peaks of surrogate data were larger than that of the real data, and P < 0.05 was considered significant. To visualize the population activity associated with protosyllables, we constructed a population raster plot by choosing 20 protosyllable renditions for which each neuron was most active. Different neurons were plotted in different colours (Fig. 2b, Extended Data Figs 1n and 9k). For all the other population raster plots associated with identified syllables, 20 random renditions were chosen for display. For all population raster plots, syllable duration from each rendition was linearly time-warped to the mean duration of the syllable. Spike times were warped by the same factor. A subset of HVC projection neurons exhibited bout-related activity: bursting before bout onsets and/or after bout offsets (Fig. 1d, e and Extended Data Fig. 2e–l). To quantify the pre-bout activity, we generated histograms aligned to bout onsets (Extended Data Fig. 2f, g) and found the peak in the histogram in a 300 ms window before bout onset. We considered a neuron to be exhibiting ‘pre-bout activity’ if the size of this peak was significant (P < 0.05) compared to peaks obtained from the shuffled surrogate histograms (identical to the procedure described earlier in the section Syllable-related neural activity). To eliminate the possibility of including syllable-related activity as bout-related activity, we did not consider a neuron to be exhibiting pre-bout activity if the neuron showed a peak in the bout-onset aligned histogram and a peak at a similar latency (less than 25 ms apart) in the syllable-onset aligned histogram. We considered a neuron to be exhibiting ‘post-bout activity’ if there was a significant peak in the bout-offset aligned histogram (Extended Data Fig. 2j, k) in a 300 ms window after bout-offset. To quantify the rhythmic neural activity of HVC projection neurons, we used four different methods: inter-burst interval, spike-train autocorrelation, spectrum of the spike train, and cepstrum of the spike train. Only spikes that were produced during singing (that is, between the onset of the first syllable and the offset of the last syllable in the bout) were used for the calculation of these measures. (1) Inter-burst interval. Intervals between burst times were calculated and the peak between 80–1,000 ms was found. (2) Spike-train autocorrelation. To quantify the second-order statistics of the firing pattern of HVC neurons, spike-train autocorrelation, expressed as a conditional firing rate61, was calculated, and the peak between 80–1,000 ms was found. The width of the centre peak indicates the width of bursts, and multiple side lobes with regular intervals indicate rhythmic bursting. (3) Spectrum of the spike train. Rhythmicity of the single-unit activity was also quantified in the frequency domain using multi-taper spectral analysis of spike trains treated as point processes62. We used the Chronux software to calculate the spectrum for the spike trains63, 64. First, bouts of singing were segmented into non-overlapping analysis windows of 1.5 s long, and then the spectrum for each window was calculated using multi-taper spectral analysis with time-bandwidth product NW = 3/2 and the number of tapers K = 2. To obtain the mean spectrum for a given neuron, spectra calculated from all the analysis windows were averaged. Finally, we found the peak in the mean spectrum within the range 2–15 Hz. (4) Cepstrum of the spike train. HVC projection neurons typically exhibited brief rhythmic bursts with precise inter-burst intervals (Fig. 1b, c). Thus, the spectrum of the spike train tended to have peaks at multiples of the fundamental frequency. To represent these burst trains that have regular intervals in a more compact way, we calculated the cepstrum (a technique commonly used in speech processing to extract the period of glottal pulses) of the spike train, defined as the inverse Fourier transform of the log spectrum65, and found the peak in the cepstrum between 80–1,000 ms. To assess the significance of the peaks in these four measures, we compared the distribution of peak amplitude obtained from the real data with that of the surrogate data obtained by shuffling the bursts times. For this shuffling procedure, we first identified all the bursts during a bout of singing as described above. We then randomly placed bursts sequentially in an interval that has the same duration as the song bout; when spikes from two bursts were closer than 30 ms, we repeated the random placement until they were spaced by more than 30 ms. Note that this randomization procedure only shuffles the burst times and preserves both the number of bursts and the ISIs within bursts. Then, all four metrics listed above were calculated by applying the same method to these surrogate spike trains. This shuffling was repeated (1,000 times for the IBI and autocorrelation, 100 times for the spectrum and cepstrum) and the P values of the peak were calculated by analysing the frequency at which the peaks from the surrogate spike trains were larger than the peak obtained from real data. A neuron was considered to exhibit ‘rhythmic’ bursting if it had significant peaks in at least two of the four metrics. The period of the rhythm was defined as the location of the largest peak of spike-train autocorrelation between 80–1,000 ms. Although many HVC projection neurons recorded in the juvenile bird exhibited rhythmic bursts, these bursts did not occur reliably on every cycle of the rhythm, but instead participated probabilistically (Fig. 2a). To quantify the degree of participation, we first extracted the protosyllables based on syllable duration (see earlier section Syllable classification and labelling) and examined the fraction of protosyllables in which at least one spike occurred (time-window from 30 ms before protosyllable onset to 10 ms after protosyllable offset). The fraction of protosyllables in which the neuron was active was obtained for all the HVC projection neurons recorded during the protosyllable stage that showed a significant rhythmic bursting (Extended Data Fig. 2p). To test whether probabilistic bursting of neurons in the protosyllable stage is coordinated across many neurons, we analysed the correlation between pairs of simultaneously recorded neurons (Fig. 2a, bottom). This analysis was restricted to pairs of neurons that were rhythmically bursting (n = 11 pairs, 3 birds). Bursting activity of each neuron was converted to a binary string corresponding to its participation in each protosyllable (for the definition of protosyllables, see earlier section Syllable classification and labelling). The activity of a neuron was assigned a ‘1’ for a protosyllable if the neuron exhibited activity in a time-window from 30 ms before protosyllable onset to 10 ms after protosyllable offset, and ‘0’ if it did not. Only activity during protosyllables was analysed to avoid including the highly variable subsong syllables, which are likely generated by circuits outside HVC27, 34. For simultaneously recorded pairs of neurons, this procedure resulted in two binary strings corresponding to the protosyllable-related activity of each neuron. We then calculated the coefficient of determination r2 by taking the square of the Pearson’s correlation coefficient r between the two binary strings. The distribution of coefficient of determination is shown in Extended Data Fig. 2q (median r2 = 0.072, 11 pairs). We also carried out a mutual information analysis to quantify whether the activity of one neuron was predictive of the set of protosyllables for which the other neuron was active. Using the same binary representation described above, we calculated the joint probability distribution describing the four possible states of activity (neither neuron spikes, neuron A spikes, neuron B spikes, both neurons spike). The mutual information was computed from this joint distribution (Extended Data Fig. 2r, median mutual information = 0.056 bits, 11 pairs). Both the correlation and mutual information were extremely low, suggesting that different projection neurons participated on relatively independent sets of protosyllables. These findings suggest that individual projection neurons participate probabilistically and largely independently in an ongoing rhythmic protosequence within HVC. We wondered whether projection neuron bursts effectively span the entire duration of juvenile song syllables, or whether bursts are highly localized to specific times, leaving other times in the syllable unrepresented22. It is clear from the syllable aligned raster plots that some syllables were completely covered by bursts (for example, Fig. 3h, syllable ‘C’), while other syllables showed some gaps in the burst coverage (for example, Fig. 4i, syllable ‘A’). To further quantify this aspect of the HVC representation during singing, we analysed the fraction of time within the syllables of juvenile birds that were ‘covered’ by the recorded projection neurons bursts (‘covered fraction’). This analysis was restricted to syllables with more than 10 associated bursts. We first determined the region of the song syllable covered by each HVC projection neuron burst. We generated a histogram of syllable -onset or -offset aligned spike times recorded from a single neuron over every recorded rendition of the song syllable. Initial identification of candidate burst events was determined by smoothing the histogram (9 ms sliding square window, 1 ms steps), and setting a threshold to define a window in which to analyse burst spikes (2 Hz for protosyllable stage birds; 10 Hz threshold for older juveniles). To eliminate low-probability spike events, we only considered bursts for which spiking activity (at least one spike) occurred in the candidate burst window on at least 25% of the renditions for that syllable. Bursts were included only if they occurred between 30 ms before syllable onset and 10 ms after syllable offset. For candidate bursts that met these criteria, all spikes occurring in the burst window were considered as contributing to that burst. Based on earlier measurements of postsynaptic currents and potentials of HVC and RA neurons66, each HVC spike in the burst window was conservatively assumed to exert a postsynaptic effect lasting no more than 5 ms. Thus, each spike in the data set was replaced with a 5 ms postsynaptic square pulse (beginning at the spike time). We considered a region of the syllable to be ‘covered’ by this burst if at least three of these post-synaptic pulses overlapped at that time within the burst, across renditions of the syllable. This procedure yielded a small ‘patch’ of time covered by the burst. The patches associated with each different neuron were combined with a logical ‘OR’ operation to determine the total coverage time of the syllable (again in a window from 30 ms before syllable onset to 10 ms after syllable offset). The covered time was divided by the duration of the syllable window to determine the covered fraction. Only syllables that had more than 10 neurons bursting within the syllable window were analysed. This criterion excluded syllables from bird 3 (shown in Extended Data Fig. 8), from which relatively few neurons were recorded. While most syllables had nearly complete burst coverage (>90%), one syllable had coverage of only 73% (Extended Data Fig. 2t), which could potentially be due to the relatively smaller number of neurons recorded in this bird. Thus, we asked whether the measured coverage is consistent with sparse sampling of the recorded bursts from a large number of uniformly placed bursts. To simulate this, we calculated the covered fraction for 1,000 surrogate data sets in which the ‘covered patches’ for each burst were randomly shuffled within the syllable. A random offset was added to the time of each patch, and a circular shift was used, allowing the patches to wrap around the edges of the syllable window. The distribution of covered fractions was determined over all shuffled surrogate data sets, and the 2.5–97.5 percentiles (95% confidence interval) of this distribution were determined (shown as vertical grey bars in Extended Data Fig. 2t). For all syllables, the observed covered fraction was consistent with that expected for random sampling from a uniform underlying distribution of burst times. To examine whether a given HVC projection neuron was active during multiple syllable types (‘shared’ neuron) or was active only during a specific syllable type (‘specific’ neuron), we first constructed a syllable-onset aligned histogram (1 ms bin, smoothed over 20 bins) for each syllable type. Spike times were linearly time warped67 to the mean duration of that syllable to reduce the trial-to-trial variability in the spike timing associated with the variation in the syllable duration. Next, we found the peak in the firing rate histogram in the interval between 30 ms before syllable onset and 10 ms after syllable offset. We visually inspected the syllable-aligned histograms, and adjusted the interval if necessary to avoid the same burst being detected twice (that is, being associated with an offset of one syllable and an onset of the next syllable). The significance of this peak was determined by comparing it with the peak size obtained from the shuffled histogram using the same method described earlier (in Syllable-related neural activity section). We defined ‘shared’ and ‘specific’ neurons in the context of a particular syllable differentiation process (for example, β and γ from bird 1 in Fig. 3; α and β from bird 2 in Fig. 4; B and D from bird 1 in Extended Data Fig. 7). ‘Specific’ neurons were defined as neurons that had a significant peak in the syllable-aligned histogram for only one syllable type, whereas ‘shared’ neurons were defined as neurons that had significant peaks for both syllable types. We took a conservative approach and only considered a neuron to be shared if the peak was significant for both syllable types. However, some neurons classified as specific had weak activity for the other syllable that did not reach significance (for example, Extended Data Fig. 6f). In other words, we believe this method likely underestimated the fraction of neurons with shared activity. Our method likely underestimated the incidence of shared neurons for another reason as well. Specifically, we defined shared and specific neurons in the context of a particular pair of syllables undergoing differentiation. For example, in a bird that exhibited hierarchical differentiation (bird 1; Extended Data Fig. 7), we saw examples of neurons that were B-specific when considering B-C differentiation but shared when considering B-D differentiation. Thus, when considering all the syllables in the motif, our definition of shared and specific neuron based on syllable pairs will underestimate the fraction of shared neurons and overestimate the fraction of specific neurons. To test whether shared neurons were active at similar latencies for multiple syllable types, we first calculated the latency of the peak in the syllable onset- or offset-aligned histograms. We then plotted the latency of the peak for one syllable against that of another syllable (Extended Data Fig. 4a–d). When a shared neuron was active for three or more syllables, two syllables associated with two highest firing rates were chosen. To quantify whether shared neurons were active at similar latencies for two syllable types, we calculated the Pearson’s correlation coefficient r between the two latencies across shared neurons, and the P value under the null hypothesis that r = 0. For the bird whose song was segmented based on the phase of the rhythm (bird 3, Extended Data Fig. 8), we asked whether bursts of shared neurons during different syllables occurred at similar phases of the rhythm. To quantify the phase of the neural activity, we first detected the burst times during singing, and for each burst, we assigned an instantaneous phase extracted from the song using the Hilbert transform (see the section on phase segmentation above). Then, the mean phase of all the bursts produced during a particular syllable type was calculated (φ , where i = 1, 2, …, 5 indicates syllables). Finally, the two syllable types were chosen for which the neuron participated most reliably, and the difference between the mean phases for these two syllables (|Δφ| = |φ − φ |, where m and n are syllable indices) was obtained (Extended Data Fig. 8i). We tested the significance of this value by comparing the value of |Δφ| against that obtained from the shuffled data where the pairing of phases were randomized across all shared neurons (Extended Data Fig. 8j; 1,000 shuffles). P values were obtained by analysing the frequency with which |Δφ| of surrogate data was smaller than that of the real data, and P < 0.05 was considered significant. To quantify the difference in the activity level for multiple syllable types in the shared neurons, we calculated the ‘bias’ defined as follows: where r is the peak firing rate in the syllable-aligned histogram for syllable i. Bias of 0 indicates equal activity level for both syllable types, whereas bias of 1 indicates exclusive activity for only one of the syllable types (Extended Data Fig. 4j). We wondered if the bursts of shared neurons were associated with different acoustic signals in the shared syllables at the time of the bursts. (An alternative possibility is that shared neurons burst only at times within the emerging syllable types when the acoustic signals are identical.) An example of a neuron analysed here is shown in Extended Data Fig. 5a (from the same data shown in Fig. 3e). This neuron bursts just after the onset of both syllables β and γ. We analysed the acoustic differences in a 0–50 ms analysis window after the burst time, but were most interested in acoustic differences in a narrower premotor window (10–40 ms), as this corresponds to the premotor latency for which one expects HVC neurons to exert an effect on vocal output29, 58, 68. For each neuron analysed, all syllables in which the neuron generated a burst were identified. The analysis was carried out for every syllable rendition on which the neuron burst, and was restricted to only those syllables. Syllables had previously been labelled by type (that is, β and γ). We first directly visualized the spectral differences between the two syllable types using a sparse contour representation69, 70, which is suitable for constructing an ‘average’ spectrogram. The analysis was carried out on the sound signal extracted from a 50 ms window after each burst. In many cases, this spectral representation revealed consistent differences between the different syllable types in this analysis window (Extended Data Fig. 5b, c). One complication is that some of the shared neurons burst before syllable onsets or immediately before syllable offsets such that the 10–40 ms window after the bursts was obscured by silent gaps (9 of 24 HVC neurons and 59 of 120 HVC neurons were obscured). These neurons were excluded from the analysis of acoustic difference. We further quantified differences in the acoustic signals by extracting time varying acoustic and spectral features in a window 0–50 ms after burst time (see subsection Definition of bursts). We used 8 acoustic features previously established to analyse birdsongs (Wiener entropy, spectral centre of gravity, spectral width, pitch, pitch goodness, sound amplitude, amplitude modulation, frequency modulation)51, 55. The 8-dimensional vector of features was calculated in 1 ms steps over the 50 ms analysis window (Extended Data Fig. 5d, e). Because each syllable was labelled, we could determine if the feature trajectories were significantly different for syllables labelled β and those labelled γ, and make this determination at every time step in the analysis window (Extended Data Fig. 5d, e; s.e.m. indicated by shaded region around mean trajectory). Rather than quantify the difference in these trajectories one feature at a time, we used Fisher’s discriminant analysis71 to project the 8-dimensional acoustic feature vector onto a single dimension that gives maximum separability between the two syllable types. The projected direction is determined independently at each time point, and the feature vectors of all syllable renditions are projected, at each time point, to yield a distribution of projected samples. For most neurons, the different syllable types produce visibly different distributions of projected samples (Extended Data Fig. 5f) indicating distinct acoustic structure. The separability of the distributions (in one dimension) of projected samples for different syllable types was quantified using the d-prime metric (d′), corresponding to the distance between the means of the distributions, normalized by the pooled variance70: Because the features evolve in time, this analysis is carried out independently at each 1 ms step in the 50 ms analysis window, and the d′ was plotted as a function of time (Extended Data Fig. 5g). Statistical significance of the d′ trajectory was assessed by randomizing the syllable labels and rerunning the d′ analysis on shuffled data sets (N = 1,000 shuffles). For each randomization, the peak value of d′ in 10-40 ms premotor window was recorded; significance threshold was set as the 95 percentile of the distribution of these peak values. A shared neuron was determined to have significant acoustic difference between the shared syllables only if the d′ trajectory remained above this significance threshold for the entire premotor window of 10–40 ms after the burst. Note that, in the simulated data, none of the 1,000 surrogate runs generated a d′ trajectory that met this stringent criterion. Results are expressed as the mean ± s.d. or s.e.m. as indicated. For χ2 tests, if the contingency table included a cell that had an expected frequency less than 5, Fisher’s exact test was used72. All tests were two-sided, and P < 0.05 was considered significant. Bonferroni correction was used to account for multiple comparisons. Figure 1f . The statistical significance of developmental changes in the fraction of HVC neurons that were syllable-aligned was assessed in two different ways: (1) Each stage was compared with the adult stage using the χ2 test followed by a post-hoc pairwise test. (2) To quantify the developmental trend in the fraction of syllable-locked neurons, we calculated Pearson’s correlation coefficient r between the binary value for each neuron (0, unlocked; 1, locked) and song stage (subsong: 1, protosyllable: 2, multi-syllables: 3, motif: 4, adult: 5). The P value was calculated under the null hypothesis that r = 0. The significance of the developmental trend for rhythmic bursting was calculated similarly. Similar results were obtained for correlation between these metrics and the age at which each neuron was recorded, rather than song stage. Figure 1g . The statistical significance of developmental changes in the period of the HVC rhythm was also assessed in two different ways: (1) Each song stage was compared with the adult stage using the Kruskal–Wallis test followed by a post-hoc pairwise test. (2) To quantify the developmental trend in the period of the HVC rhythm, we calculated Pearson’s correlation coefficient r between burst period and song stage. Similar results were obtained for correlation between burst period and the age at which each neuron was recorded. Figure 2c . The Wilcoxon rank-sum test was used to test whether the median of the syllable-onset aligned latency distribution was different between subsong and protosyllable stages. Figures 3g, h and 4h, i . To test whether the fraction shared neurons differed between early and late stages of syllable differentiation, we used the χ2 test on a 2 × 2 contingency table (shared/specific, early/late). Regarding across all birds, to calculate whether the fraction of shared neurons differed between early and late stages of syllable differentiation over all birds (n = 5 syllable pairs in 3 birds), we used the Cochran–Mantel–Haenszel test for repeated tests of independence73. Extended Data Fig. 1a. To quantify the relation between song stage and age, we calculated Spearman’s rank correlation coefficient ρ and the P value under the null hypothesis that ρ = 0. Extended Data Fig. 1c. We computed the statistical significance of developmental changes in burst width (top) and firing rate during bursts (bottom) by using the Kruskal–Wallis test followed by a post-hoc pairwise test to compare each stage with the adult stage. Extended Data Fig. 2m–o. To test whether fraction of syllable-locked neurons (Extended Data Fig. 2m), fraction of rhythmic neurons (Extended Data Fig. 2n), and period of HVC rhythm (Extended Data Fig. 2o) significantly differed between HVC and HVC , we used χ2 test for all the pairwise comparisons with Bonferroni correction for multiple comparisons. Extended Data Fig. 4a–d. To calculate the relation between latencies of bursts associated with shared neurons, we calculated the Pearson’s correlation coefficient r together with the P value under the null hypothesis that r = 0. Extended Data Fig. 5m, n. To test whether the mean d′ metric was different between HVC and HVC , we used the Wilcoxon rank-sum test. Only neurons with d′ trajectories that were significant (continuously from 10–40 ms) were included in this comparison. Code used to simulate the model is available as Supplementary Information. To illustrate a potential mechanism of chain splitting, we chose to implement the model as simply as possible. We modelled neurons as binary units and simulated their activity in discrete time steps44; at each time step (10 ms), the ith neuron either bursts (x = 1) or is silent (x = 0). A network of 100 binary neurons is recurrently connected in an all-to-all manner, with W representing the synaptic strength from presynaptic neuron j to postsynaptic neuron i. Self-excitation is prevented by setting W = 0 for all i at all times44. During learning, the strength of each synapse is constrained to be within the interval [0, w ], while the total incoming and outgoing weights of each neuron are both constrained by the “soft bound” W =m*w where m represents a target number of saturated synapses per neuron44 (see section Synaptic plasticity rule for details). Note that w represents a hard maximum weight of each individual synapse, while W represents a soft maximum total synaptic input or output of any one neuron. Synaptic weights are initialized with random uniform distribution such that each neuron receives, on average, its maximum allowable total input, W . The activity of each neuron in the network was determined in two steps: calculating the net feedforward input that comes from the previous time step; then determining whether that is enough to overcome the recurrent inhibition in the current time step. First, the net feedforward input to the ith neuron at time step t, , was calculated by summing the excitation, feedforward inhibition, neural adaptation, and external inputs: where [z] indicates a rectification (equal to z if z > 0 and 0 otherwise). is the excitatory input from network activity on the previous time step. is a global feedforward inhibitory input44, where β sets the strength of this feedforward inhibition. is an adaptation term44 where α is the strength of adaptation, and y is a low-pass filtered record of recent activity in x with time constant τ = 40 ms; that is ; B (t) is the external input to neuron i at time t. For seed neurons, this term consists of training inputs (see section on Seed neurons). For non-seed neurons, it consists of random inputs with probability p = 0.01 in each time step and size W /10. Finally, θ is a threshold term used to reduce the excitability of seed neurons, making them less responsive to recurrent input than are other neurons in the network. For seed neurons, θ = 10 and for non-seed neurons, θ = 0. Including this term improves robustness of the training procedure by eliminating occasional situations in which seed neuron activity may be dominated by recurrent rather than external inputs. In these cases, external inputs may fail to exert proper control of network activity. Second, we determined whether the ith neuron will burst or not at time step t by examining whether the net feedforward input, , exceeds the recurrent inhibition, AI_rec(t). We implemented recurrent inhibition by estimating the total input to the network at time t: and feeding it back to all the neurons. Parameter γ sets the strength of the recurrent inhibition. We assume that this recurrent inhibition operates on a fast time scale48 (that is, faster than the duration of a burst). Thus, the final output of the ith neuron at time t becomes: where Θ[z] is the Heaviside step function (equal to 1 if z > 0 and 0 otherwise). To induce splitting, γ was gradually stepped up to γ following a sigmoid with time constant τ and inflection point t : A subset of neurons was designated as seed neurons, which received external training inputs used to shape network activity during learning43, 45. The external training inputs activate seed neurons at syllable onsets, reflecting the observed onset-related bursts of HVC neurons during the subsong stage (Fig. 1a). The pattern of these inputs was adjusted in different stages of learning, and each strategy of syllable learning was implemented by different patterns of seed neuron training inputs. Alternating differentiation (Fig. 5a–e). Ten neurons were designated as seed neurons and received strong external input (W ) to drive network activity. In the subsong stage, seed neurons were driven (by external inputs) synchronously and randomly with probability 0.1 in each time step corresponding to the random occurrence of syllable onsets in subsong27, 34. This was done only to visualize network activity; no learning was implemented at the subsong stage. During the protosyllable stage, seed neurons were driven synchronously and rhythmically with a period T = 100 ms. The protosyllable stage consisted of 500 iterations of 10 pulses each. To initiate chain splitting, the seed neurons were divided into two groups and each group was driven on alternate cycles. The splitting stage consisted of 2,000 iterations of 5 pulses in each group of seed neurons (1 s total per iteration, as in the protosyllable stage). Motif strategy (Extended Data Fig. 10e–h). This was implemented in a similar manner as alternating differentiation, except that 9 seed neurons were used, and for the splitting stage, seed neurons were divided into 3 groups of 3 neurons, each driven on every third cycle. Bout-onset differentiation (Extended Data Fig. 10a-d). Seed neurons were divided into two groups: 5 bout-onset seed neurons and 5 protosyllable seed neurons. At all learning stages, external inputs were organized into bouts consisting of four separate input pulses, and bout-onset seed neurons were driven at the beginning of each bout. Then, 30 ms later, protosyllable seed neurons were driven three times with an interval of T = 100 ms. In the protosyllable stage, inputs to all seed neurons were of strength W . In the splitting stage, the input to protosyllable seed neurons was decreased to W /10. This allowed neurons in the bout-onset chain to suppress, through fast recurrent inhibition, the activity of protosyllable seed neurons during bout-onset syllables. Each iteration of the simulation was 5 s long, consisting of 10 bouts, described directly above, with random inter-bout intervals. The protosyllable stage consisted of 100 iterations, and the splitting stage consisted of 500 iterations. Bout-onset syllable formation (Extended Data Fig. 10i–k). Input to seed neurons was set high (2.5*W ), and maintained at this high level throughout development. This prevented protosyllable seed neurons from being inhibited by neurons in the bout-onset chain. Furthermore, strong external input to the protosyllable seed neurons terminated activity in the bout-onset chain through fast recurrent inhibition, thus preventing further growth of the bout-onset chain, as occurs in bout-onset differentiation. As in bout-onset differentiation, each iteration of the simulation was 5 s long, consisting of 10 bouts with random inter-bout intervals. The protosyllable stage consisted of 100 iterations, and the splitting stage consisted of 500 iterations. As in previous models43, 44, we hypothesized two plasticity rules in our model: Hebbian spike-timing dependent plasticity (STDP) to drive sequence formation74, 75, and heterosynaptic long term depression (hLTD) to introduce competition between synapses of a given neuron43, 44. STDP is governed by the antisymmetric plasticity rule with a short temporal window (one burst duration): where the constant η sets the learning rate. hLTD limits the total strength of weights for neuron i, and the summed weight limit rule for incoming weights is given by: and for outgoing weights from neuron j: At each time step, total change in synapse weight is given by the combination of STDP and hLTD: where ε sets the relative strength of hLTD. In our implementation of the subsong stage, there was no learning. Subsong model parameters were: β = 0.115, α = 30, η = 0, ε = 0, γ = 0.01. After subsong, learning progressed in two stages: the protosyllable stage and the splitting stage. Parameters that remained constant over development were: β = 0.115, α = 30, η = 0.025, ε = 0.2. To induce chain splitting, w , the maximum allowed strength of any synapse, was increased from 1 to 2, m was decreased from 10 to 5, and γ was increased from 0.01 to 0.18 following a sigmoid with time constant τ = 200 iterations and inflection point t = 500 iterations into the splitting stage. No change in parameters occurred before the chain-splitting stage. Parameters that remained constant over development were: β = 0.13, α = 30, η = 0.05, ε = 0.14. To induce chain splitting, w was increased from 1 to 2, m was decreased from 5 to 2.5, and γ was increased from 0.01 to 0.04 following a sigmoid with time constant τ = 200 iterations and inflection point t = 250 iterations into the splitting stage. Parameters that remained constant over development were: β = 0.115, α = 30, η = 0.025, ε = 0.2. To induce chain splitting, w was increased from 1 to 2, m was decreased from 9 to 3, and γ was increased from 0.01 to 0.18 following a sigmoid with time constant τ = 200 iterations and inflection point t = 500 iterations into the splitting stage. Parameters that remained constant over development were: β = 0.13, α = 30, η = 0.05, ε = 0.15. To induce chain splitting, w was increased from 1 to 2, m was decreased from 5 to 2.5, and γ was increased from 0.01 to 0.05 following a sigmoid with time constant τ = 200 iterations and inflection point t = 250 iterations into the splitting stage. Neurons were classified as participating in a syllable type if the syllable onset-aligned histogram exhibited a peak that passed a threshold criterion. The criteria were chosen to include neurons where the histogram peak exceeded 90% of surrogate histogram peaks. Surrogate histograms were generated by placing one burst at a random latency in each syllable. (For example, in the protosyllable stage, the above criterion was found to be equivalent to having 5 bursts at the same latency in a bout of 10 protosyllables.) During the splitting phase, neurons were classified as shared if they participated in both syllable types, and specific if they participated in only one syllable type. We visualized network activity in two ways: network diagrams, and raster plots of population activity (for example, Fig. 5a–d top and bottom panels, respectively). In both cases, we only included neurons that participated in at least one of the syllable types (see earlier section Shared and specific neurons for participation criteria). Network diagrams. Neurons are sorted along the x axis based on their relative latencies. Neurons are sorted along the y axis based on the relative strength of their synaptic input from specific neurons (or seed neurons) of each type (red or blue). Lines between neurons correspond to feedforward synaptic weights, and darker lines indicate stronger synaptic weights. For clarity of plotting, only the strongest six outgoing and strongest nine incoming weights are plotted for each neuron. Population raster plots. Neurons are sorted from top to bottom according to their latency. Groups of seed neurons are indicated by magenta arrows. Shared neurons are plotted at the top and specific neurons are plotted below. As for network diagrams, neurons that did not reliably participate in at least one syllable type were excluded. Further details for Fig. 5a–d. Panels show network diagrams and raster plots at four different stages. Figure 5a shows subsong stage (before learning), Fig. 5b shows end of protosyllable stage (iteration 500), Fig. 5c shows early chain splitting stage (iteration 992), Fig. 5d shows late chain-splitting stage (iteration 2,500). Code used to simulate the model is available as Supplementary Information.
News Article | November 13, 2015
The Commonwealth Bank wants to ramp up its lending to renewable energy businesses, and has signalled it believes Australia should ultimately re-intstate a carbon price. Group executive Kelly Bayer Rosmarin, who runs CBA's institutional bank, on Friday said CBA had "unlimited" appetite for lending to renewable energy, though she also highlighted the challenges banks faced in financing the sector.
Schaefer B.E.,Louisiana State University |
Pagnotta A.,Louisiana State University |
Lacluyze A.P.,University of North Carolina at Chapel Hill |
Reichart D.E.,University of North Carolina at Chapel Hill |
And 34 more authors.
Astrophysical Journal | Year: 2011
The eruption of the recurrent nova U Scorpii on 2010 January 28 is now the all-time best observed nova event. We report 36,776 magnitudes throughout its 67 day eruption, for an average of one measure every 2.6 minutes. This unique and unprecedented coverage is the first time that a nova has had any substantial amount of fast photometry. With this, two new phenomena have been discovered: the fast flares in the early light curve seen from days 9-15 (which have no proposed explanation) and the optical dips seen out of eclipse from days 41-61 (likely caused by raised rims of the accretion disk occulting the bright inner regions of the disk as seen over specific orbital phases). The expanding shell and wind cleared enough from days 12-15 so that the inner binary system became visible, resulting in the sudden onset of eclipses and the turn-on of the supersoft X-ray source. On day15, a strong asymmetry in the out-of-eclipse light points to the existence of the accretion stream. The normal optical flickering restarts on day 24.5. For days 15-26, eclipse mapping shows that the optical source is spherically symmetric with a radius of 4.1 R⊙. For days 26-41, the optical light is coming from a rim-bright disk of radius 3.4 R ⊙. For days 41-67, the optical source is a center-bright disk of radius 2.2 R⊙. Throughout the eruption, the colors remain essentially constant. We present 12 eclipse times during eruption plus five just after the eruption. © 2011. The American Astronomical Society. All rights reserved. Source
Timed natural matings were used for all experiments, where noon of the next day after the vaginal plugs of mated females were identified was scored as E0.5. Animal studies were authorized by a UK Home Office Project License and carried out in a Home Office-designated facility. No statistical methods were used to predetermine sample size. ∆PE-Oct3/4-GFP (GOF-GFP)31, 32, Prdm1-GFP and Prdm1−/− ES cell lines were established previously14, 19, 22, 23. Inducible Sox2-knockout (2CG2) ES cell line was a gift from H. Niwa27. All ES cell lines were maintained in naive ‘ground state’33 condition; that is, in N2B27 medium (R&D) with 2i (PD0325901, 1 μM; CHIR99021, 3 μM; Stemgent), LIF and 1% KnockOut Serum Replacement (KSR; Life Technologies) on fibronectin-coated dishes (16.7 μg ml−1; Millipore). Medium was changed daily. ES cell colonies were passaged by dissociating with TrypLE (Life Technologies). Oct3/4, Sox2, Nanog, Prdm1, Prdm14 and Brachyury complementary DNAs were cloned from mouse cDNA pool. cDNAs were inserted into PiggyBac-based doxycycline (Dox)-inducible vectors (a gift from H. Niwa). These vectors were transfected using Lipofectamine 2000 (Life Technologies) into ES cells together with a pPyCAG-PBase vector and a pPBCAG-rtTAIRESNeor vector, which harbours a neomycin resistance gene. After 1 week of neomycin (80 μg ml−1; Life Technologies) selection, pooled or single clones were used for experiments. To induce transgene expression, various concentrations of Dox (Sigma-Aldrich) were added to the media. EpiLCs and PGCLCs were induced as described previously5. Transgenes were induced by addition of Dox at day 0 of PGCLC induction. 100 ng ml−1 Dox was used in experiments shown in Figs 3f, 4e, g and Extended Data Figs 3b, c, 7g–i, 8a–c. 200 ng ml−1 Dox was used in Fig. 4b–d, f. 700 ng ml−1 of Dox was used in all other experiments. PGCLCs were induced as described in the manuscript. For inhibition of the BMP–SMAD pathway, noggin (200 ng ml−1; R&D) was added to the media at day 0 of PGCLC induction. For inhibition of WNT signalling, XAV939 (1 μM; Sigma-Aldrich) was added to the media. Day-1 or day-2 EpiLCs were transferred into GMEM 15%KSR 2i/LIF with or without Dox in monolayer culture. In addition, day-1 or day-2 EpiLCs were aggregated in low-cell-binding U-bottom-shaped 96-well plates (Thermo Scientific) (1,000–2,000 cells per well) in PGCLC induction media (GMEM with l-glutamine (Life Technologies), 15% KSR (Life Technologies), 1× MEM NEAA (Life Technologies), 1× sodium pyruvate (Life Technologies), 1× 2-mercaptoethanol (Life Technologies), 1× penicillin/streptomycin (Life Technologies)) and Dox. The medium was replaced daily. After 3 days, the GFP reporter signal was analysed with a fluorescence microscope and via FACS analysis. RNA was collected from pooled cells for qRT–PCR. Day-4 aggregates were dissociated with TrypLE and plated on mitomycin C-treated mouse embryonic fibroblast (MEF) feeder cells with PGC selection medium (DMEM with l-glutamine (Life Technologies), 15% fetal bovine serum (FBS; Sigma-Aldrich)), LIF, 15 ng ml−1 bFGF, 30 ng ml−1 SCF (R&D) and 2 μM all trans-retinoic acid (Sigma-Aldrich). Retinoic acid promotes germ cell self-renewal while promoting differentiation of ES cells20, 21. The media was replaced daily. After 5 days, proliferating GFP+ cells were dissociated with TrypLE and plated on fibronectin-coated dishes with ESC medium (N2B27 with 2i/LIF). PGCLCs were dissociated with TrypLE, washed with DMEM containing 10% FBS and resuspended with 1×PBS containing 0.1% BSA. Large clumps of cells were removed using a cell strainer (BD Biosciences). The cells were analysed and sorted on flow cytometers (FACS Calibur, BD Biosciences; MoFlo high speed cell sorter, Beckman Coulter; S3 cell sorter, Biorad). Total RNAs from ES cells, EpiLCs and FACS-sorted cells were extracted using the RNeasy Mini Kit (Qiagen) or Picopure RNA Isolation Kit (Life Technologies). The total RNAs were reverse transcribed by the Quantitect Reverse Transcription Kit (QIAGEN). The first-strand cDNAs were used for RT–qPCR analysis with SYBR Green PCR reagent (Sigma-Aldrich). The primer sequences used for the qRT–PCR are listed in Supplementary Table 1. Student’s t-test was used to test for significance. ES cells and day-4 PGCLCs were dissociated and sorted with a MoFlo high-speed cell sorter (Beckman Coulter). Total RNAs were extracted using the RNeasy Mini Kit (QIAGEN). Complementary RNA (cRNA) generation, quality control, hybridization and data analysis were performed by Cambridge Genomic Services at the University of Cambridge. Raw intensity values from Illumina MouseWG-6 v.2.0 expression beadchip microarrays were pre-processed with the Bioconductor lumi and preprocessCore packages (http://www.bioconductor.org): Probes that were not detected in at least one sample were removed, Variance stabilization transformation (VST) was applied, and samples were quantile-normalized. Differential expression was evaluated with the Bioconductor limma package. Comparison with published microarray data (Extended Data Fig. 3j). Our data set was assayed on an Illumina MouseWG-6 v.2.0 expression beadchip, the data set from ref. 5 was assayed on the Affymetrix Mouse Genome 430 2.0 Array platform. We therefore quantile-normalized the data sets to ensure that the data sets span comparable ranges of expression values. PCA was performed on the centre-scaled expression values, where systematic differences between platforms are mainly captured by the first principal component. Day-3, day-4 and day-6 aggregates were fixed with 2% or 4% paraformaldehyde for 20 min at room temperature or for 2 h at 4 °C. Fixed aggregates were washed several times in PBS and transferred into 10% sucrose/PBS (2 h), 20% sucrose/PBS (2 h) and finally into OCT embedding matrix (overnight; CellPath). Next day, cell aggregates were embedded in OCT in tissue moulds and stored at −80 °C. A Leica Cryostat CM3050S was used to cut the OCT blocks in 6–8-μm-thick sections, which were collected on SuperFrost Plus slides (VWR). For immunofluorescence staining, the slides were washed with PBS, permeabilized with PBS/0.1–1% Triton X-100 and then incubated with primary antibodies in permeabilization buffer including 5% donkey serum (Sigma-Aldrich) overnight at 4 °C. Next day, slides were washed with PBS and incubated with secondary antibodies in permeabilization buffer for 2 h at room temperature, washed with PBS, incubated with 4′,6-diamidino-2-phenylindole (DAPI) in PBS for 15–30 min, and mounted using Vectashield Mounting Medium (VECTOR Labs). Images were acquired using a Leica SP5 or SP8 confocal microscope. For 5hmC stainings, it was required to perform an additional antigen retrieval step before incubation with primary antibodies: slides with sections were transferred into TE buffer, pH 8, at ~95 °C and microwaved at very low power for 45 min. The following primary antibodies were used: mouse anti-OCT3/4 (1:100, BD Biosciences, O50808), rat anti-BLIMP1 (1:50, eBioscience, clone 6D3, 14-5963), rabbit anti-AP-2γ (1:250, SantaCruz, sc-8977), rabbit anti-PRDM14 (1:250, a gift from D. Reinberg), rabbit anti-DAZL (1:500, Abcam, ab34139), mouse anti-H3K9me2 (1:250, Abcam, ab1220 and 1:500, Millipore, 07-441), rabbit anti-H3K27me3 (1:500, Millipore, 07-449), rabbit anti-TET1 (1:500, Millipore, 09-872), rabbit anti-5hmC (1:500, Active Motif, 39791), goat anti-KLF4 (1:100, R&D, AF3158), rabbit anti-H3S10ph (1:500, Millipore, 06-5770), mouse anti-γH2A.X (1:250, Millipore, 05-636), rat anti-GFP (1:500, Nakalai Tesque, GF090R). Alexa Fluor488 and 568 were used as secondary antibodies (1:500, Life Technologies). All quantifications were preformed using Fiji34. The DAPI, H3S10ph and γH2A.X channels were processed by applying a Gaussian Blur (H3S10ph staining: DAPI/H3S10ph: σ 0.5/1.1; DAPI/ γH2A.X: σ 1.0/1.5) to reduce noise. The images were then binarized using the Otsu thresholding algorithm and holes were filled before the total signal area was measured. In day-6 Prdm1−/− plus Dox aggregates, many cells underwent cell death. Therefore, nuclei with bright discrete spots of DAPI signal, which indicates chromatin condensation, were excluded from the analysis. The diameter of ~10 cells was measured and used to calculate the average area of one cell to estimate the number of cells in the field of view (DAPI+ area/area of one cell). For all other quantifications on a single-cell level, we developed ‘Object Scan’, which is an object mapping and analysis plugin for Fiji that combines advanced functions with a user-friendly interface. Images are processed with a choice of feature enhancement algorithms, objects are identified by patch sampling to detect intensity edges based on the local energy gradient, and the generated two-dimensional masks are clustered in three dimensions to define the final object map for analysis. We used Object Scan to carry out DoG processing and contained signal analysis using the DAPI channel for object mapping, watershed segmentation, a scan radius of one and the following channel specific settings: edge gradient = 10, estimated object radius = 9 μm. The results were scale normalized (X − X /X − X ) to the range 0 to 1 for comparison. Student’s t-test was used to test for significance. The Object Scan plugin is available from this link: http://www.gurdon.cam.ac.uk/stafflinks/downloadspublic/imaging-plugins. Low cell number ChIP-qPCR was performed as previously described35. 3 × 105 cells per ChIP were fixed in 1% formaldehyde (room temperature, 10 min), quenched with 1 vol. of 250 mM glycine (room temperature, 5 min), and rinsed with chilled TBSE buffer (20 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA) twice before storage at −80 °C. After thawing the cells on ice, fixed cells were lysed with 100 μl 1% SDS lysis buffer (50 mM Tris-HCl pH 8, 10 mM EDTA, 1% SDS, Roche protease inhibitor cocktail; on ice, 5 min) and then centrifuged (2,000 r.p.m., 10 min). Pellet was resuspended in 100 μl of dilution buffer (16.7 mM Tris-HCl, pH 8, 167 mM NaCl, 1.2 mM EDTA, 1.1% Triton X-100, 0.01% SDS, Roche protease inhibitor cocktail). Samples were sonicated nine times (30-s pulses with 30-s break interval) using the Bioruptor water bath sonicator (Diagenode). Chromatin extracts were then precleared with Dynal Magnetic Beads (Invitrogen) (4 °C, 1 h) followed by centrifugation (2,000 r.p.m., 30 min). Supernatant (precleared chromatin) was immunoprecipitated overnight with Dynal Magnetic Beads coupled with anti-NANOG antibody (1 μg per ChIP, Cosmo Bio Co., RCAB0001P) or normal rabbit serum (1 μg per ChIP). On the next day, beads were washed (nutate in wash buffer for 5 min at 4 °C) in low-salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8.0, 150 mM NaCl), high-salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8.0, 300 mM NaCl) and LiCl buffer (0.25 M LiCl, 1% NP400, 1% Na deoxycholate, 1 mM EDTA, 10 mM Tris-HCl, pH 8.0), for a total of three washes. Following an additional wash in TE, elution was performed in a PCR machine (68 °C, 10 min). After digesting and reverse crosslinking (with Proteinase K at 42 °C for 2 h and 68 °C for 6 h) DNA was purified (phenol-chloroform extraction) and used for qPCR analysis. For the negative control region, we used the Snai3 locus as described previously36. Student’s t-test was used to test for significance. The same protocol was used for the SOX2 ChIP with some deviations. Day-2 EpiLCs were aggregated in low-binding plates for 6 h in the presence of 200 ng ml−1 of Dox before collection. 5 × 106 ES cells and EpiLCs, respectively, were fixed and processed as described earlier. Samples were sonicated 20 times (30-s pulses with 30-s break interval) using a Bioruptor water bath sonicator (Diagenode). Samples were divided for immunoprecipitations with SOX2 antibody (10 μg per ChIP, Santa Cruz, sc-17320 X) or normal rabbit IgG (10 μg per ChIP, Santa Cruz, sc-2027 X) as a negative control. Beads were washed with low-salt buffer, and twice with high-salt buffer for 10 min each. The beads were rinsed in TE, resuspended in Proteinase K digestion buffer (20 mM HEPES, 1 mM EDTA, 0.5% SDS) with 2 μl of 10 mg ml−1 Proteinase K and incubated for 15 min at 50 °C. In parallel, 2 μl of 10 mg ml−1 Proteinase K was added to the saved input samples. Three microlitres 5 M NaCl was added to the supernatants and the input samples. To reverse the crosslinks, samples were incubated at 42 °C for 2 h and 68 °C overnight. Next day, the DNA was purified using Agencourt Ampure XP beads (Beckman Coulter) according to the manufacturer’s instructions. The purified DNA was used for qPCR analysis. For the negative control region, we used the Snai3 locus. Student’s t-test was used to test for significance. The primer sequences used for RT–qPCRs are listed in Supplementary Table 1. The NANOG ChIP for subsequent sequencing was performed as described earlier with some deviations. Day-1 or day-2 EpiLCs were aggregated in low-binding plates for 3 h in the presence of 200 ng ml−1 of Dox. ES cells and EpiLCs were fixed and processed as described earlier. 3 × 106 fixed cells were lysed with 1 ml 1% SDS lysis buffer and then centrifuged (2,000 r.p.m., 15 min). Nuclear fraction was resuspended in 0.9 ml of dilution buffer. Samples were sonicated ten times (30-s pulses with 30-s break interval) using a Bioruptor water bath sonicator (Diagenode). Immunoprecipitations were performed with anti-NANOG antibody (2 μg per ChIP, Cosmo Bio Co., RCAB0001P). After elution, samples were digested with Proteinase K and reverse crosslinked for 6 h at 68 °C. Twelve nanograms of purified DNA was used for library preparation using Ovation Ultralow DR Multiplex System (Nugen). Once prepared, library was size selected and sequenced using HiSeq2000 with single-end 50 nucleotides read length. ChIP-seq reads were aligned with the bwa aligner (http://bio-bwa.sourceforge.net) to the mouse reference genome (GRCm38/mm10). Peaks were called with MACS (version 2.1.0; https://github.com/taoliu/MACS) and visualized using the Integrative Genomics Viewer (https://www.broadinstitute.org/igv/). Peak regions from two biological replicates were intersected using bedops (http://bedops.readthedocs.org). Overlapping peak regions with peak summits within <50 nucleotides distance in both replicates were retained. Peak regions from the three cell types were merged. Differences in ChIP-seq read intensities on peak regions were evaluated by using diffReps (http://code.google.com/p/diffreps) and MACS (macs2 bdgdiff). High-confidence sets of differentially bound regions that were detected by both methods were selected for further analysis by applying the following thresholds for diffReps: pValue <0.001 and abs(log2FC) > 1. Previously published H3K27ac ChIP-seq data sets9, 30 were aligned to the mouse reference genome in a similar manner as described earlier, and H3K27ac enrichment (log(ChIP/input) values were determined on NANOG peak regions. High-confidence MACS peaks, for which the distance of the peak summits in both replicates was <50 nucleotides, were selected. De novo motifs were determined with HOMER (http://homer.salk.edu/homer) in the 2,000 top-enriched peaks in ES cells, day-1 and day-2 EpiLCs for both repeat-masked and repeat-unmasked regions within ± 50 nucleotides of the peak summit. Genomic regions containing putative enhancers of Prdm1 and Prdm14, as well as a negative control region depleted of enhancer signatures, were amplified from mouse E14 ES-cell genomic DNA. These regions were cloned into a PiggyBAC-based firefly luciferase reporter plasmid upstream of a minimal TK promoter. Stable luciferase reporter GOF-GFP ES cell lines, which can overexpress Nanog, Nanog/Sox2 or Brachyury upon Dox addition, were established. Cell pellets were collected from ES cells cultured in N2B27 2i/LIF, day-2 EpiLCs and EpiLCs after PGCLC induction ±Dox at 12/24 h. Luciferase assays were performed with the ONE-Glo Luciferase Assay System (Promega). Protein concentration in each lysate was quantified by Pierce 660 nm Protein Assay (Thermo Scientific). Relative luciferase activities were obtained by dividing luciferase activity by protein concentration in each sample. ES-cell clones carrying both the Nanog transgene and a CAG monomeric Kusabira-orange (mKO) fluorescence reporter were selected by neomycin (Sigma-Aldrich) and zeocine (Life Technologies). Day-4 PGCLCs were induced from day-2 EpiLCs with Nanog and used for derivation of EGCLCs. For ES cells or day-4 PGCLC injections, GOF-GFP ES cells were co-transfected with a vector, which enabled inducible expression of Nanog and constitutive expression of Venus, a variant of eGFP. For day-4 PGCLCs, after induction of PGCLCs with Nanog, cells were stained with PE-conjugated-CD61 antibody (1:10, Biolegend, 104308) and Alexa660-conjugated-SSEA-1 antibody (2.5 μl per 105 cells, eBioscience, clone eBioMC-480, 50-8813) according to the manufacturer’s instructions. Double-positive PGCLC cells were collected by using a S3 cell sorter (Biorad). Embryos for chimaera experiments were obtained from CBA/C57BL/6 F1 crossed with C57BL/6 mice. Blind tests or randomization methods were not used. The sex of embryos was not determined. Manipulations of embryos were performed as described previously37. Briefly, five cells were injected into a morula, which were subsequently cultured in KSOM (Millipore). On the following day, the embryos were transferred into the uteri of pseudopregnant mice. All embryos were analysed 1 week after embryo transfer, which corresponded to E9.5. The CRISPR–Cas9 system was used to generate Nanog-knockout ES cells. Guide RNAs (gRNAs) targeting exon 1 of the Nanog gene were cloned into pX330 (Addgene)38. One microgram of this plasmid was transfected with a pPyCAG-monomeric Kusabira Orange-IRES-Pac plasmid. Transfected cells were selected by puromycin (1 μg ml−1) for 2 days. Clonal Nanog-knockout ES cell lines were established and mutations of Nanog alleles were confirmed by qPCR, western blotting and DNA sequencing. Subsequently, pPBhCMV*1-Nanog-pA plasmid was transfected into those lines with pPyCAG-PBase and pPBhCMV*1-rtTA-IRESNeor to generate Nanog-knockout ES cell lines carrying a Dox-inducible Nanog transgene. Loss of Nanog affected the growth of ES cells. Thus, these cell lines were maintained in N2B27 2i/LIF with a low dose of Dox (100 ng ml−1). gRNA sequences are listed in Supplementary Table 1. 5 × 104 cells were lysed in lysis buffer (50 mM Tris-HCl (pH 8.0), 1% SDS, 10 mM EDTA). Protein concentration was measured by Bicinchoninic Acid Kit (Sigma-Aldrich). The protein amount was adjusted among samples, then 4 × Laemmli buffer was added. Samples were boiled at 95 °C for 5 min. Proteins were separated on 10% acrylamide gels, blotted on Immobilon-P transfer membrane (Millipore). The membrane was blocked with 5% skimmed milk and incubated with primary antibodies: anti-NANOG (1:500, mouse IgG, eBioscience, clone eBioMLC-51, 14-5761), anti-SOX2 (1:500, rabbit IgG, Cell Signaling, 2748), anti-α-tubulin (1:1,000, mouse IgG, Sigma-Aldrich, clone DM1A, T9026). Primary antibodies were detected on X-ray film with anti-rabbit or anti-mouse IgG conjugated with horseradish peroxidase (Dako) followed by detection using Western Detection System (GE Healthcare). For gel source data, see Supplementary Fig. 1. pPBhCMV*1-Nanog-pA, pPBCAG-rtTA-IRESNeor and pPyCAG-PBase were transfected into the Sox2-conditional-knockout ES cell line (2CG2)27. After 1 week of neomycin selection (80 μg ml−1), pooled cells were used for the subsequent experiments. Dexamethasone-inducible Sox2-knockout and Dox-inducible Nanog expression were confirmed by qPCR and western blotting.
Carter P.J.,University of Warwick |
Steeghs D.,University of Warwick |
de Miguel E.,University of Huelva |
Goff W.,CBA |
And 12 more authors.
Monthly Notices of the Royal Astronomical Society | Year: 2013
We present time-resolved spectroscopy and photometry of the dwarf nova SBSS 1108+574, obtained during the 2012 outburst. Its quiescent spectrum is unusually rich in helium, showing broad, double-peaked emission lines from the accretion disc. We measure a line flux ratio He I 5875/Hα = 0.81 ± 0.04, a much higher ratio than typically observed in cataclysmic variable stars (CVs). The outburst spectrum shows hydrogen and helium in absorption, with weak emission of Hα and He I 6678, as well as strong He II emission. From our photometry, we find the superhump period to be 56.34 ± 0.18 min, in agreement with the previously published result. The spectroscopic period, derived from the radial velocities of the emission lines, is found to be 55.3 ± 0.8 min, consistent with a previously identified photometric orbital period, and significantly below the normal CV period minimum. This indicates that the donor in SBSS 1108+574 is highly evolved. The superhump excess derived from our photometry implies a mass ratio of q = 0.086 ± 0.014. Our spectroscopy reveals a grazing eclipse of the large outbursting disc. As the disc is significantly larger during outburst, it is unlikely that an eclipse will be detectable in quiescence. The relatively high accretion rate implied by the detection of outbursts, together with the large mass ratio, suggests that SBSS 1108+574 is still evolving towards its period minimum. © 2013 The Authors. Published by Oxford University Press on behalf of the Royal Astronomical Society. Source