A developmental framework linking neurogenesis and circuit formation in the Drosophila CNS

  1. Brandon Mark
  2. Sen-Lin Lai
  3. Aref Arzan Zarin
  4. Laurina Manning
  5. Ashok Litwin-Kumar
  6. Albert Cardona
  7. James W. Truman
  8. Chris Q. Doe

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This paper sought to assess the relationship between developmental lineage and connectivity. It relies on detailed EM reconstructions and the knowledge of complete neuroblast lineages, thus correlating wiring with lineage. Through genetic manipulations of Notch function, it also correlates developmental programs with wiring. The conclusion is important and provides a well described cellular and genetic system for linking the developmental program of a cell to its connection specificity. It provides a framework for considering how to study these questions in other regions of the Drosophila brain and can be extended to the study of more complex mammalian systems where a similar neuroblast-lineage strategy generates different neuron types.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 and Reviewer #3 agreed to share their names with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

The mechanisms specifying neuronal diversity are well-characterized, yet it remains unclear how or if these mechanisms regulate neural circuit assembly. To address this, we mapped the developmental origin of 160 interneurons from seven bilateral neural progenitors (neuroblasts), and identify them in a synapse-scale TEM reconstruction of the Drosophila larval CNS. We find that lineages concurrently build the sensory and motor neuropils by generating sensory and motor hemilineages in a Notch-dependent manner. Neurons in a hemilineage share common synaptic targeting within the neuropil, which is further refined based on neuronal temporal identity. Connectome analysis shows that hemilineage-temporal cohorts share common connectivity. Finally, we show that proximity alone cannot explain the observed connectivity structure, suggesting hemilineage/temporal identity confers an added layer of specificity. Thus, we demonstrate that the mechanisms specifying neuronal diversity also govern circuit formation and function, and that these principles are broadly applicable throughout the nervous system.

Article activity feed

  1. eLife

    Author Response:

    Reviewer #1:

    The authors sought to assess the relationship between developmental lineage and connectivity.

    This is a tour de force. It relies on detailed EM reconstructions, knowledge of complete neuroblast lineages thus correlating wiring with lineage, and through genetic manipulations of N gene function correlates developmental programs with wiring. The conclusion is important and provides a well described cellular and genetic system for linking the developmental program of a cell to its connection specificity. It provides a framework for considering how to study these questions in other regions of Drosophila and can be extended to the study of more complex mammalian systems where a similar neuroblast-lineage strategy generates different neuron types.

    There are no major weakness.

    This is an excellent study and, in my opinion, is ready to publish in its current form.

    We appreciate this comment!

    Reviewer #2:

    The conclusions of this paper are mostly well supported by data, however, there are several points that should be discussed further in the manuscript:

    1. The authors state that overexpression of Notchintra transforms Notch OFF neurons into Notch ON neurons. However, since this decision happens at the level of the GMC, wouldn't be more correct to say that Notch OFF neurons were not produced and only Notch ON neurons were generated? Moreover, the authors state that the Notchintra overexpression phenotypes are due to hemilineage transformation rather than to death of Notch OFF neurons, by providing the total neuronal number in both experimental conditions using NB5-2 lineage. I think this statement is too much of a generalization when only one NB lineage has been analyzed and should be addressed in more lineages to claim this as a general mechanism. Moreover, the opposite hypothesis could have also been tested to make the argument stronger: Would depletion of Notch in GMCs make all neurons in a lineage target the ventral neuropil domain?

    We agree, and now provide cell counts for WT and Notch-intra in all four lineages (5-2, 7-1, 7-4, and 1-2) in the text. In all cases, the number of neurons in wild type and Notch-intra lineages are not significantly different, supporting the Notch OFF to Notch ON transformation. We don't say that Notch-OFF neurons are missing, because there is no loss of neurons from the lineage, but rather the neurons that would have been Notch-OFF in wild type are now duplicating the Notch-ON neurons. Regarding presenting the opposite transformation, we tried to do it with misexpressing UAS-numb, but were unable to get the expected positive control phenotype in which all five Eve+ U neurons are transformed to Eve-negative siblings (Skeath and Doe, 1998). Thus, we were not able to do lineage-specific Notch inhibition. Unfortunately, we can’t use whole embryo N or N pathway mutants, as has been done before (Skeath and Doe, 1998), because they have massive disruption in the CNS that obscures lineage specific axon phenotypes.

    1. Temporal cohorts described in this work are an approximation to neuronal temporal identity. The authors validate the correlation of early and late temporal cohorts to the expression of the temporal TFs Hb and Cas (Fig 4G). Given the resolution of the TEM dataset and the existence of specific NBs and neuronal drivers for the neurons studied, a correlation between the 4 temporal cohorts presented in this work and the 4 temporal TFs Hb, Kr, Pdm and Cas expressed by these neurons could have been possible and would have presented a more comprehensive view of the relationship between tTF expression and neurite and synapse localization. Does temporal cohort between lineages (cortex neurite length) mean expression of the same temporal TF? For example: would mid-early neurons in different lineages express the same temporal factor?

    Excellent question! We show that radial position is a proxy for temporal identity, but the precise relationship of Hb, Kr, Pdm, and Cas expressing neurons to the four radial “bins” we describe remains unknown. In fact, a graduate student is doing these experiments by generating MCFO single neuron clones in newly hatched larvae (the stage of the TEM volume) and staining with Hb, Kr, and Cas temporal transcription factors (it is impossible to so this with Pdm because neurons lose expression at stage 15). This will be many months of work and probably over a thousand MCFO+ neurons to analyze, and we feel it is beyond the scope of the paper -- although very important and very interesting! Plus, we are still limited in lab time due to University of Oregon covid restrictions.

    Since shared temporal identity between different lineages on its own does not confer shared neuronal projections, but shared temporal cohort hemilineage does: Does this mean that the expression of a given temporal TF and/or neuronal birth order does not play a role in this shared connectivity? Please clarify these ideas in the text.

    We have tried to clarify this in the text. Whereas temporal identity alone has no detectable role in generating common synapse localization or connectivity, it does have some role in the context of hemilineage identity. That is, hemilineage temporal cohorts have more shared synapse localization and connectivity than either temporal or hemilineage identity alone. See Figure 6 for synapse localization, and Figure 7 for connectivity data.

    1. Although the authors claim so, it is not convincing that the role of spatial patterning in neuronal connectivity has been assessed in this work, since the authors do not present an obvious correlation between specific connectivity features (morphology, axon or synapses localization) and the position of a given NB in the VNC. This should be clarified in the text.

    Great point! We agree that spatial patterning was not directly tested in our manuscript, thank you for pointing this out. Our claim that spatial patterning is involved is simply based on the idea that lineages (and thus hemilineages) are more related to one another than neurons from other hemilineages suggesting that the identity of the parent neuroblast plays some role. You make the excellent point that we did not look at the relationship of projections from all NBs in a “row” or “column” within the NB array. That analysis would potentially reveal a role for spatial factors in determining neuron projections. Unfortunately, we have a very limited set of neurons from any one row or column, not enough to make claims about direct relationships between row or column identity and targeting/connectivity.

    Reviewer #3:

    Specific comments:

    1. Figure 1; page 3: The authors refer to the "striking" similarity between EM reconstructions and GFP filled clones and yet there are clear differences in some of the clones in the extent and localization of arborization. This may be in part technical but almost certainly also reflects inter individual differences in single neuron morphology. Since EM reconstructions presumably come for, one animal, the use of GFP clones allows the authors to map the degree of variation between clones and it would be interesting for them to show this.

    That is an interesting point. Elegant work from Tzumin Lee and Jim Truman have shown that clones from larval neuroblasts are very similar, and our qualitative findings support this conclusion. Thus, it would be a quite minor advance for us to quantify clonal similarity in embryonic neuroblasts. Plus, since the number of neurons in a clone varies slightly, we would have to count neuron numbers per clone and only compare those with identical neuron numbers, which is possible but time-consuming. Then there are the covid restrictions which make it difficult to rapidly generate new clones to increase the number with identical neurons. All in all, we decided that the benefit of answering this question was not worth the cost of performing it, and that other experiments were a higher priority in our limited research time. We have toned down the language to remove the word “striking” in the Introduction.

    1. Figures 2 and 4; pages 3-5: Along the same lines as above, the authors make categorical statements about the mapping of arbors to dorsal and ventral regions of the nerve cord and correlate that to hemilineage identity. Again, there is clear mixing in almost all neuroblast lineages, that seems to range from 15-30% as a rough estimate, and perhaps a bit more dorsally than ventrally, which the authors do not comment on (except to say it's "mostly non-overlapping"). This is a pity because they obviously have the tools to do so quantitatively and the information is already there in their data.

    Yes, good point – there is some overlap in most lineages for both axon/dendrite targeting (Figure 2) and synapse targeting (Figure 4). We now quantify the synapse similarity and observe that hemilineage-related neurons have much greater synapse similarity than they have with their sister hemilineage. The non-overlapping relationship between hemilineages is somewhat obscured by the simple posterior view shown in Figures 2 and 4, so we add a new figure (Figure 4 – supplement 2) that shows hemilineage synapse targeting in all three axis: A/P, M/L, and D/V. This makes it possible to see the true relationship.

    1. The analysis of Notch activity in hemilineages is excellent and very interesting, as is the new tool they develop. However, the analysis lacks loss of Notch function data and where and when Notch signaling is required to segregate the connectivity space (i.e. in neurons or in precursors such as Nbs and GMCs). Is this a binary fate specification mechanism or lateral inhibition among competing neurons? What about Notch activity manipulation in single neurons? If the authors wish to draw strong conclusions about the role of Notch in segregating target space and its relation to hemilineage identity, these experiments are essential. Alternatively, drawing subtler conclusions and acknowledging these caveats would be very welcome.

    Great point about the possible role of non-canonical Notch signaling in post-mitotic neurons (PMID: 22608692). We do not have the tools to perform lineage-specific, axon-specific removal of Notch protein. In theory we could do single neuron MARCM experiments, but these are extremely difficult due to the perdurance of the Gal80 protein, which would prevent us from assaying in newly hatched larvae. We add a Discussion section addressing the unresolved issue of post-mitotic neuron Notch function: “Another point to consider is the potential role of Notch in post-mitotic neurons (Crowner et al., 2003), as our experiments generated Notch-intra misexpression in both new-born sibling neurons as well as mature post-mitotic neurons. Future work manipulating Notch levels specifically in mature post-mitotic neurons undergoing process outgrowth will be needed to identify the role of Notch in mature neurons, if any.”

    1. Figure 7; Page 7: The authors state that 75% of hemilineage neurons correlated by temporal identity are separated by 2 synapses or less, suggesting greater connectivity than expected. How are these data normalized? What is the expected connectivity between neurons that are less related along these two developmental axes?

    Thanks for the question, which helped us change the text for clarity. The quantifications in Figure 7 actually do compare connectivity between unrelated neurons. Thus, we have changed “random” to “unrelated” in the text and figure legends. Additionally, the methods for this analysis were obviously not clear enough, so we have updated them with this text below:

    Path Length Analysis:

    We computed the pairwise path length between all hemilineages as well as all sensory and motor neurons in A1 in the undirected connectivity graph. We found that neurons that are unrelated by developmental grouping had an average path length greater than that of neurons related by hemilineage. Additionally, we found that the average path length for neurons related by hemilineage alone had an average path length greater than that of neurons in hemilineagetemporal-cohorts. For this analysis, unrelated neurons were defined as neurons that were in the same D/V axis (i.e. dorsal to dorsal and ventral to ventral) and same hemisegment (left or right), but not in the same hemilineage. Hemilineage comparisons were neurons in the same hemilineage, but not in the same temporal cohort. Significance was determined with a two-sample KS test on the empirical distributions of pairwise path lengths.

    Independent of path length, we also calculated connectivity similarity between related neurons in Figure 8. Similarity here was defined as the cosine of the angle between the input or output vectors of each neuron. Similarity by this metric was also found to be greater for developmentally related neurons. Finally, we added this line to Figure 7 legend to clarify normalization: “Frequency corresponds to the fraction of pairwise distances observed for each group.”

    1. Figure 8; page 7 and discussion: The authors conclude that the combination between temporal identity and hemilineage identity predicts connectivity beyond what would be predicted by spatial proximity alone. This conclusion is problematic at least two levels. First, practically what really matters for proximity is proximity during the time in development when synapses are forming between neuronal pairs, not proximity at the end in the final pattern.

    This is a good point that we need to clarify, although we note that synaptic connectivity is not a "one and done" in the embryo, but rather a continuous process that extends from the late embryo into the third larval instar ("Conserved neural circuit structure across Drosophila larval development revealed by comparative connectomics" by Gerhard, Andrade, Fetter, Cardona, and Schneider-Mizell, eLife 2017).

    Nevertheless, we now add the following additional text to the Results and to the Discussion. To the Results: “Interestingly, even neurons with the highest observed levels of overlap were not always connected (Figure 8A''). Thus, proximity alone can't explain the observed connectivity, consistent with a role for hemilineage-temporal cohorts providing increased synaptic specificity. Of course, our assays are in newly hatched larvae, and it is likely that dendritic arbors are more widely distributed during circuit establishment in the late embryo (Valdes-Aleman et al., 2021), yet only a specific region of the neuropil is targeted by larval hatching, which suggests the initial broad dendrite targeting is not sufficient to establish connectivity to many neurons contacted by these early dendrites, again arguing against a simple proximity mechanism.” To the Discussion: “Our results strongly suggest that hemilineage identity and temporal identity act combinatorially to allow small pools of neurons to target pre- and postsynapses to highly precise regions of the neuropil, thereby restricting synaptic partner choice. Yet precise neuropil targeting is not sufficient to explain connectivity, as many similarly positioned axons and dendrites fail to form connections (Figure 8C), despite active synapse addition throughout larval life (Gerhard et al., 2017).”

    Second, conceptually, opposing spatio-temporal mechanisms with proximity-based bias for connectivity makes no sense because that's exactly what spatio-temporal mechanisms achieve: getting neurons to the same space at the same time so connectivity can happen. At any rate, drawing strong conclusions about where and when neurons meet to form (or not form) synapses requires live imaging and absent that authors should refrain from making such a string statement about what their excellent correlative dataset means.

    Yes, spatiotemporal mechanisms get axons (or dendrites) to precise neuropil domains, but that does not invariably generate connectivity. What is interesting is that hemilineage-temporal cohorts share more connectivity than predicted by proximity alone. Thus, proximity is necessary but not sufficient for proper connectivity. An additional mechanism is in play, and our data suggests that is due to the neuron's hemilineage-temporal identity. We agree that our data are correlative – shared development correlates with shared connectivity – so we have moved any suggestion of possible mechanism from the Results to the Discussion. We agree this is an important change that will increase manuscript accuracy, and also provide a clear future direction for mechanistic experiments. Thanks for helping us focus the paper better.

  2. eLife

    Reviewer #3 (Public Review):

    Mark and colleagues set out to examine the relationship between neuronal targeting and connectivity and the developmental history of neurons. Specifically, the authors examine if, how and to what extent hemilineage identity combined with temporal birth order can explain neuronal connectivity. To this end they use the fly larval nerve cord with its EM-level resolution of connectivity and prior knowledge about hemilineage identity and birth order as a model.

    General comments:

    The manuscript represents a comprehensive, thorough and deep analysis of the system. The descriptive elements are outstanding, and the analysis of how Notch activity correlates with hemilineage targeting is of great interest. While understanding the relationship between connectivity diagrams and developmental history, including the role of Notch signaling, is not new, the scale of the analysis presented here is key because it allows - in principle - the drawing of general conclusions. The main "weakness" of the manuscript is not in the work itself but rather some of the key conclusions drawn from the data which often somewhat beyond what the data alone would support, especially in terms of the developmental mechanisms involved in establishing connectivity. With one partial exception (Notch gain of function experiments), the work essentially represents (very important) correlation analysis between the various parameters. While the authors are of course free to interpret their data as per their own views and biases, they do need to either tone down their, often categorical, statements and soften their conclusions, or perform further analysis to examine whether some of the stronger conclusions they draw are justified.

    Specific comments:

    1. Figure 1; page 3: The authors refer to the "striking" similarity between EM reconstructions and GFP filled clones and yet there are clear differences in some of the clones in the extent and localization of arborization. This may be in part technical but almost certainly also reflects inter individual differences in single neuron morphology. Since EM reconstructions presumably come for, one animal, the use of GFP clones allows the authors to map the degree of variation between clones and it would be interesting for them to show this.

    2. Figures 2 and 4; pages 3-5: Along the same lines as above, the authors make categorical statements about the mapping of arbors to dorsal and ventral regions of the nerve cord and correlate that to hemilineage identity. Again, there is clear mixing in almost all neuroblast lineages, that seems to range from 15-30% as a rough estimate, and perhaps a bit more dorsally than ventrally, which the authors do not comment on (except to say it's "mostly non-overlapping"). This is a pity because they obviously have the tools to do so quantitatively and the information is already there in their data.

    3. The analysis of Notch activity in hemilineages is excellent and very interesting, as is the new tool they develop. However, the analysis lacks loss of Notch function data and where and when Notch signaling is required to segregate the connectivity space (i.e. in neurons or in precursors such as Nbs and GMCs). Is this a binary fate specification mechanism or lateral inhibition among competing neurons? What about Notch activity manipulation in single neurons? If the authors wish to draw strong conclusions about the role of Notch in segregating target space and its relation to hemilineage identity, these experiments are essential. Alternatively, drawing subtler conclusions and acknowledging these caveats would be very welcome.

    4. Figure 7; Page 7: The authors state that 75% of hemilineage neurons correlated by temporal identity are separated by 2 synapses or less, suggesting greater connectivity than expected. How are these data normalized? What is the expected connectivity between neurons that are less related along these two developmental axes?

    5. Figure 8; page 7 and discussion: The authors conclude that the combination between temporal identity and hemilineage identity predicts connectivity beyond what would be predicted by spatial proximity alone. This conclusion is problematic at least two levels. First, practically what really matters for proximity is proximity during the time in development when synapses are forming between neuronal pairs, not proximity at the end in the final pattern. Second, conceptually, opposing spatio-temporal mechanisms with proximity-based bias for connectivity makes no sense because that's exactly what spatio-temporal mechanisms achieve: getting neurons to the same space at the same time so connectivity can happen. At any rate, drawing strong conclusions about where and when neurons meet to form (or not form) synapses requires live imaging and absent that authors should refrain from making such a string statement about what their excellent correlative dataset means.

  3. eLife

    Reviewer #2 (Public Review):

    This work presents a comprehensive characterization of seven different neuronal lineages and their connections in the Drosophila CNS. By making use of a previously generated TEM reconstruction of the first instar larval CNS, the authors map de developmental origin of 160 interneurons, providing an unvaluable framework to address the question of how specification mechanisms in neural stem cells, such as temporal patterning, and Notch status of the neurons, correlate with different aspects of neuronal connectivity. The authors show that most NB lineages produce two morphologically distinct hemilineages, each one targeting either the ventral or dorsal VNC neuropil domain in a Notch-dependent manner, which allows the concurrent building of these circuits in a similar number. Importantly, they show that Notch activity is sufficient to target neurons to the dorsal neuropil domain and that hemilineage-related neurons share similar synapse localization. Furthermore, by measuring the cortex neurite length of these neurons, the authors establish neuronal cell body radial position as a proxy of neuronal birth order and assign different temporal cohorts to these neurons. Importantly, they show that neurons that share a hemillineage temporal cohort identity have more similar synaptic positions and shared connectivity than neurons that share hemilineage identity, providing an additional level of partner specificity than that provided by hemillineage alone. They further show that the observed shared connectivity between hemilineages and hemilineage temporal cohorts cannot be explained by proximity alone, further validating the observations presented in this work.

    This work is of great value for the field of Developmental Neurobiology as it provides an initial understanding of the link between neuronal specification mechanisms and circuit formation during development. By mapping specific neuronal lineages in a serial section TEM reconstruction, the authors analyze neuronal connectivity with single synapse resolution, allowing a precise characterization of neurite localization and synaptic specificity, which are not offered in most of the works published in the field. The availability (and additional generation in this work) of drivers to label specific NB lineages and hemilineages on the VNC combined with TEM, presents an unvaluable resource to study circuit formation during development. The use of this high-quality framework in the future will continue deepening our understanding of how specification mechanisms in neural stem cells instruct circuit formation and connectivity, and the molecular mechanisms underlying these processes.

    The conclusions of this paper are mostly well supported by data, however, there are several points that should be discussed further in the manuscript:

    1. The authors state that overexpression of Notchintra transforms Notch OFF neurons into Notch ON neurons. However, since this decision happens at the level of the GMC, wouldn't be more correct to say that Notch OFF neurons were not produced and only Notch ON neurons were generated? Moreover, the authors state that the Notchintra overexpression phenotypes are due to hemilineage transformation rather than to death of Notch OFF neurons, by providing the total neuronal number in both experimental conditions using NB5-2 lineage. I think this statement is too much of a generalization when only one NB lineage has been analyzed and should be addressed in more lineages to claim this as a general mechanism. Moreover, the opposite hypothesis could have also been tested to make the argument stronger: Would depletion of Notch in GMCs make all neurons in a lineage target the ventral neuropil domain?

    2. Temporal cohorts described in this work are an approximation to neuronal temporal identity. The authors validate the correlation of early and late temporal cohorts to the expression of the temporal TFs Hb and Cas (Fig 4G). Given the resolution of the TEM dataset and the existence of specific NBs and neuronal drivers for the neurons studied, a correlation between the 4 temporal cohorts presented in this work and the 4 temporal TFs Hb, Kr, Pdm and Cas expressed by these neurons could have been possible and would have presented a more comprehensive view of the relationship between tTF expression and neurite and synapse localization. Does temporal cohort between lineages (cortex neurite length) mean expression of the same temporal TF? For example: would mid-early neurons in different lineages express the same temporal factor? Since shared temporal identity between different lineages on its own does not confer shared neuronal projections, but shared temporal cohort hemilineage does: Does this mean that the expression of a given temporal TF and/or neuronal birth order does not play a role in this shared connectivity? Please clarify these ideas in the text.

    3. Although the authors claim so, it is not convincing that the role of spatial patterning in neuronal connectivity has been assessed in this work, since the authors do not present an obvious correlation between specific connectivity features (morphology, axon or synapses localization) and the position of a given NB in the VNC. This should be clarified in the text.

  4. eLife

    Reviewer #1 (Public Review):

    The authors sought to assess the relationship between developmental lineage and connectivity.

    This is a tour de force. It relies on detailed EM reconstructions, knowledge of complete neuroblast lineages thus correlating wiring with lineage, and through genetic manipulations of N gene function correlates developmental programs with wiring. The conclusion is important and provides a well described cellular and genetic system for linking the developmental program of a cell to its connection specificity. It provides a framework for considering how to study these questions in other regions of Drosophila and can be extended to the study of more complex mammalian systems where a similar neuroblast-lineage strategy generates different neuron types.

    There are no major weakness.

    This is an excellent study and, in my opinion, is ready to publish in its current form.

  5. eLife

    Evaluation Summary:

    This paper sought to assess the relationship between developmental lineage and connectivity. It relies on detailed EM reconstructions and the knowledge of complete neuroblast lineages, thus correlating wiring with lineage. Through genetic manipulations of Notch function, it also correlates developmental programs with wiring. The conclusion is important and provides a well described cellular and genetic system for linking the developmental program of a cell to its connection specificity. It provides a framework for considering how to study these questions in other regions of the Drosophila brain and can be extended to the study of more complex mammalian systems where a similar neuroblast-lineage strategy generates different neuron types.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 and Reviewer #3 agreed to share their names with the authors.)

  6. Version published to 10.1101/617936v3 on bioRxiv
  7. Version published to 10.1101/617936v2 on bioRxiv
  8. Version published to 10.1101/617936v1 on bioRxiv