Morphometric analysis of lungfish endocasts elucidates early dipnoan palaeoneurological evolution

Curation statements for this article:
  • Curated by eLife

    eLife logo

    Evaluation Summary:

    Clement and colleagues describe and illustrate the endocasts of six Palaeozoic lungfish genera from superb 3D fossil material, which are very informative for the understanding of brain evolution of lungfishes, the extant sister group to land vertebrates. Rendering important anatomical details regarding brain evolution in lungfishes, and sarcopterygians in general, this work will be of broad interest to zoologists, including vertebrate paleontologists and neuroanatomists.

    (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 #, Reviewer #2 and Reviewer #3 agreed to share their name with the authors.)

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

The lobe-finned fish, lungfish (Dipnoi, Sarcoptergii), have persisted for ~400 million years from the Devonian Period to present day. The evolution of their dermal skull and dentition is relatively well understood, but this is not the case for the central nervous system. While the brain has poor preservation potential and is not currently known in any fossil lungfish, substantial indirect information about it and associated structures (e.g. labyrinths) can be obtained from the cranial endocast. However, before the recent development of X-ray tomography as a palaeontological tool, these endocasts could not be studied non-destructively, and few detailed studies were undertaken. Here, we describe and illustrate the endocasts of six Palaeozoic lungfish from tomographic scans. We combine these with six previously described digital lungfish endocasts (4 fossil and 2 recent taxa) into a 12-taxon dataset for multivariate morphometric analysis using 17 variables. We find that the olfactory region is more highly plastic than the hindbrain, and undergoes significant elongation in several taxa. Further, while the semicircular canals covary as an integrated module, the utriculus and sacculus vary independently of each other. Functional interpretation suggests that olfaction has remained a dominant sense throughout lungfish evolution, and changes in the labyrinth may potentially reflect a change from nektonic to near-shore environmental niches. Phylogenetic implications show that endocranial form fails to support monophyly of the ‘chirodipterids’. Those with elongated crania similarly fail to form a distinct clade, suggesting these two paraphyletic groups have converged towards either head elongation or truncation driven by non-phylogenetic constraints.

Article activity feed

  1. Author Response

    Reviewer #2 (Public Review):

    First, I want to congratulate the author team on this manuscript, which I read with great pleasure. I think this will be a fine addition to the literature!

    The present MS by Clement et al. provides a comprehensive overview of the brain shapes of lungfishes. Besides previously known/described brain endocasts, the work includes models and descriptions of previously undescribed taxa. Notably, all CT data are deposited online following best practices when working with digital anatomy. The specimen sample is impressive, especially as the sampled material is housed in museum all over the world. Although the sample size may seem numerically low (12 taxa), this actually is a comprehensive sample of fossil (and extant) lungfishes in terms of what's preserved in the first place.

    The study at hand has several goals: (1) The description of lungfish brains for taxa that were previously undescribed; (2) the quantification of aspects of brain shape using morphometric measurements; (3) the characterization of brain shape evolution of lungfishes using exploratory methods that ordinate morphometric measurements into a morphospace.

    The provided 3D data and descriptions will serve as valuable comparisons in future lungfish work. This type of data is imperial for palaeontological studies in general, and the anatomical information will be extremely valuable in the future. For example, anatomical characters related to brain architecture have been shown to be informative about phylogeny in the past, and the presented data may inform future phylogenetic studies. The quantification of brain shape via (largely linear) measurements is relatively simplistic, and can thus only detect gross trends in brain shape evolution among lungfishes. The authors describe several such trends - such as high variation in the olfactory brain region in comparison to other parts of the brain. The results and interpretations drawn from the authors are supported by their data, and the approach taken is valid, even if more sophisticated shape quantification methods (e.g. 3D landmarking) and analytical methods (e.g. explicit phylogenetic comparative methods) are available, which could provide additional insights in the future.

    We agree with Reviewer #2 that 3D geometric morphometrics could have provided more sophisticated analytical methods. However, geometric morphometrics has some limitations with regard to the type of data that we analysed: (1) low sample size and (2) missing/incomplete data. In order to have a comprehensive coverage of the brain shape, it would have required to have numerous landmarks (and semilandmarks) to represent the complexity of brain shape.

    First, our sample size (12 taxa) is low (although it is an impressive sample size when considering the type of data). Although there are no universal rule concerning the ratio “number of specimens / number of landmarks” (Zelditch et al., 2012), ideally the sample size must be from two to three times the number of landmarks. Thus, with a sample size of 12 we could have used ca. 4-6 landmarks which is very limited to describe complex shapes. In addition, in order to use geometric morphometrics (2D or 3D), the landmarks should be present on all the specimens. Because of the partial completeness of the studied fossils, the brain endocasts are not uniformly known for each species. Incomplete and deformed specimens prompt the removal of potential landmarks for analyses. Even using right-left reflexion of the endocasts, most specimens do not share all neurocranial information.

    We agree with Reviewer #2 that a phylogenetic PCA could have provided interesting analytical perspectives. Phylogenetic PCA are available on standard PCA, it is uncertain that it can be used on Bayesian PCA and InDaPCA (this method has been published very recently, and we haven’t found much literature about it). However, we did not find an adaptation of phylogenetic PCA to the BPCA nor the InDaPCA; we even contacted Liam Revell, who created the phylogenetic PCA, about this issue.

    The presented results and interpretations in this regard must be seen as a preliminary assessment of lungfish brain evolution, but it is clearly written and generally well performed.

    A potential shortcoming of the paper is the lack of explicit hypothesis testing, which is not problematic per se, but puts limits on the conclusions the authors can draw from their data.

    We decided to address the issues using exploratory methods rather than testing hypotheses. It is a more conservative approach, since it is the first quantitative analysis of dipnoan endocasts. Future analyses, will be able to formulate hypotheses based on our interpretation of our exploratory approach. We hope to stimulate such hypotheses testing, when in the future further dipnoans will be added; however, one has to remember that ossified neurocrania are known in Devonian dipnoans and one partially ossified neurocranium in a Carboniferous, the remaining dipnoans have cartilaginous neurocrania which limit the sample size from which endocast data could be gathered.

    For example, the authors state that different anatomical parts of the labyrinth (particularly, the utricle with respect to the semicircular canals or saccule) may show modular dissociation from other labyrinth modules, based on the polarity of eigenvalue signs of the PCA analysis. I think this is fine as a first approximation, but of course there are explicit statistical tools available to test for modularity/integration, such as two-block partial least squares regression analysis (Rohlf & Corti 2000, Syst. Biol.). I don't see the lack of usage of such methods as problematic, because you cannot do everything in one paper, and the authors remain careful in their interpretation.

    We agree with Reviewer #2 that different geometric morphometrics methods have been developed to look at variational modularity; one of the co-authors (RC) has been publishing a few papers on patterns of morphological integration and modularity in fishes (see Larouche, Cloutier & Zelditch, 2015, Evol. Biol.; Lehoux & Cloutier, 2015, J. Exp. Zool. Mol. Dev. Evol.; Larouche, Zelditch & Cloutier, 2018, Sci. Rep.). Interesting a priori hypotheses of brain modules could have been formulated and tested for modularity using for example Covariance Ratio (CR) and distance matrix approach. But still the low sample size and the incompleteness of the data are major constrains to test modularity. We would however endeavour to use such methods in future work as more complete material becomes available.

    It may be advisable, however, to add the odd sentence or statement about how some findings are preliminary or hypothesized, and that these should receive further treatment and testing using other methods in the future. I think this approach is actually very rewarding, because then you can inspire future work by outlining outstanding research problems that arise from the new data presented herein.

    We have now included an additional sentence early in the Discussion section stating: “We acknowledge that our investigation of lungfish brain evolution as elucidated from morphometric analysis of cranial endocasts is still preliminary in several respects. We hope that our study can inspire future work on the neural evolution of both fossil and extant lungfish.”

    In the following, I comment on a few aspects of the manuscripts. These represent instances where I had additional thoughts or ideas on how to slightly improve various aspects of the manuscript.

    1. Presentation of PCA results

    The authors provide several PCA analyses (preliminary analyses on partial matrices, BPCA, InDaPCA), and are very explicit about the procedures in general. For instance, I appreciate they explicitely state using correlation matrices for PCA analyses due to the usage of different measurement units among their data.

    Visually, the BPCA and InDaPCA are presented in figures 2 and 3, whereas the preliminary partial matrix PCAs are only reported as supplementary figures. While I don't object to any of this, I find the sequence of information given in the results section suboptimal.

    The figures have now been substantially reorganised to include more within the main body text and not as Supplementary Information, and we hope that this improves the sequence of information within the manuscript.

    The authors start by discussing the partial matrix analyses, although none of these analyses are visually/graphically depicted in the main text figures, and although their results do not seem to be of real importance for the narrative of the discussion. The other two PCA analyses actually are presented afterwards and separately, but they convey some common signals, particularly that the major source of variation seems to be a decreasing olfactory angle with increasing olfactory length, and a scaling relationship between all linear measurements (which all have the same eigenvector signs on the first PC axis). I wonder if an alternative way of presenting the PCA results would be better for this particular MS. For example, the authors could give "first level observations" first ("PCA analyses agree in X,Y,Y"), and then move to second order observations ("Morphospace of BPCA has some interesting taxon distribution with regard to chirodipterids"; "InDaPCA axis projections continuously retrieve clustering of specific variables"). I suspect this would shorten the text somewhat and could serve as a clearer articulation of the take home messages?

    Accordingly with Reviewer #2, we have now provided “first level” observations based on the standard PCA. We added some further comments on the species distribution in the morphospaces.

    1. Selection of PC axes for interpretation

    You describe how you use the broken-stick method to decide how many PC axes are retained for the interpretation of results, which I agree is a good procedure. However, I have a few questions regarding this. First, in line 331 (description of InDaPCA) you state that the first three axes are non-trivial "based on the screeplot" - which got me confused because it sounds a bit like eyeballing off the screeplot. Have you used the broken stick method for all your PCA analyses?

    Originally, we used both screeplot and broken-stick method, however, we are now solely using the broken stick method to determine the number of non-trivial axes. We agree with Reviewer #2 that this method is more rigorous than the scree plot. Our choice is greatly inspired by the studies of Jackson (1993, Ecology) and PeresNeto et al. (2005, Computational Statistics & Data Analysis). We have now edited the text so that our methods are clearer (and removed the text relating to the screeplot such as “based on the screeplot…”).

    The second question relates to the results of the broken stick method, which I did not find reported. Unless I am mistaken, for the xth axis, the method sums the fractions of 1/i (whereby i = x..n; n = number of axes), and divides this number by n to get a value of expected variation per axis. This number is then compared with the actual value of variance explained by the axis. So for the 1st of 17 axes, the broken-stick expectation is = (1 + 1/2 + .. + 1/17) / 17. If you apply this to your BPCA, the third axis' value (i.e., (1/3 + ... + 1/17)/17) is 0.114, which is smaller than the reported 0.120 that PC3 explains. Thus, following the broken stick method, PC3 does explain more variation that expected (and should thus be retained, contra your comment in line 311 which refers to two non-trivial axes)?

    We thank Reviewer #2 for the insightful evaluation of our paper who took the time to validate each step of our analyses. Effectively, we agree with Reviewer #2 that based on the broken stick method the third axis in nontrivial. The value for the third axis is 1,0531310. Thus, we are presenting these results as well as discussing the three PCA projections (axis 1 versus axis 2, axis 2 versus axis 3, axis 1 versus axis 3).

    Related to this potential issue is the presentation of the BPCA results in Fig. 2: You present loadings of three PC axes, although only the first two are considered in morphospace bi-plots and although the text also mentions only two non-trival axes. If the third axis is indeed non-trivial, then the loading-presentation could be retained in the figure, but then the authors should consider showing a PC1 vs. PC3 plot in addition to the currently presented biplot showing the first and second axis only. If the third axis indeed is trivial, as currently suggested by the text, then showing the loadings is unnecessary.

    We consider showing a biplot of PC1 vs PC3 unnecessary as those shown (PC1 vs PC2) already account for 83.4% of the variation captured. We have edited these figures so that the loadings related to PC3 have also now been omitted.

    It would be great if you clarify the usage/application of the broken stick method for all your PCAs. An easy way to report the results may be the add a row to each of your PCA loading tables in the supplements, in which you divide the actual value of variation explained by the value expected under the broken stick method - this way, all axes which explain more variation than expected by the stick method have values larger than 1, and axes which explain less have values lower than 1.

    We have taken this suggestion from Reviewer #2 on board and have now recalculated all values for the brokenstick method for each analysis; we also provide broken-stick values in their respective loading tables in the SI.

    1. Missing commentary on allometry

    In basically all PCA analyses, the first PC axis seems to be dominated by allometric size effects, given that all linear measurements have the same eigenvalue signs. The authors do acknowledge this (lines 314-316; 335-336), but offer no further comment on size effects/allometry.

    We agree that normally the first axis represents variation related mainly to size changes and shape changes related to size (allometry). However, we are reluctant to assume that our first axis corresponds to evolutionary allometry. Among others, Klingenberg & Zimmermann (1992) and Klingenberg (1996) used standard PCA (or multi-group PCA) to disentangle evolutionary and ontogenetic allometry (as well as static allometry) mainly by analysing multiple specimens for each group (or species) in order to have a better repartition of the covariance. Since our sample is limited to 12 species, and that they are all represented by a single specimen (except for Dipterus), it would be difficult to clearly discriminate variation associated to allometry. Even in a case of ontogenetic allometry, a sample size of 12 would have been limited to unambiguously conclude any variation.

    For example, it would be interesting to see how the linear measurements scale with overall head size. Similarly, the authors note that the semicircular canal measurements covary strongly, as do the utricle and saccule height/length measurements (paragraph line 346). Basically, it seems that the semicircular canal measurements scale with one another: as one gets bigger, so gets the other. It is interesting that the utricle does not seem to follow the same scaling pattern as the saccule and semicircular canals, and it would be good to hear if the authors think that there is a functional implication for this. Increases in utricular/saccular/semicircular canal sizes are usually explained by increased sensitivity - so is an increased utricular size a compensatory development to decreased semicircular canal+saccule size to retain an overall level of sensitivity, or does it maybe related to a relative change of importance of the specific functions, e.g. increased importance of linear accelerations in the horizontal plane with simultaneous decrease of importance of angular and vertical accelerations?

    We thank Reviewer 2 for this suggestion about overall head size scaling - endocast measurements. Our original study design also included measurements of dermal skulls, but we omitted this from the final version as the material available was far too incomplete to be able to conduct meaningful analyses. It is a topic of future study that some of us (AC, RC) have already discussed as a potential future project to be investigated.
    With respect to the functional implications of the modular dissociation of the labyrinths, we have expanded the final paragraph of the “implications for sensory abilities” within the Discussion, and similarly added the sentence “However, we acknowledge that it is difficult to determine if increased relative utricular size results from greater reliance of sensitivity in the horizontal plane alone, or if it expands to compensate for e.g. relative stagnation of the sacculus + semicircular canals in some way. Further studies, such as investigation of neuronal densities in extant lungfish labyrinths, may potentially in part clarify this uncertainty in future.”

    1. Labyrinth size

    With the above mentioned utricular exception, labyrinth size measurements particularly on the semicircular canals seem to imply that there is a relative consistent scaling relationship between the canals. When one canal gets larger, so do the others, perhaps thereby retaining canal symmetry across different absolute labyrinth sizes. Labyrinth size in tetrapods is often interpreted in relation to body size/mass or head size (e.g. Melville Jones & Spells 1963, Proc. R. Soc. Lond. Biol. Sci.; Spoor & Zonneveldt 1998, Yearb. Phys. Anthr.; Spoor et al. 2002, Nature; Spoor et al. 2007, PNAS; Bronzati et al. 2021, Curr. Biol.), as deviations from the expected labyrinth size per head size indicate increased or decreased relative labyrinth sensitivities. Large relative head sizes of birds and (within) mammals have generally been interpreted as indicative of "active" or "agile" behaviour, although doubt has been casted on these relationships recently (e.g., Bronzati et al. 2021). Increased sampling of relative labyrinth size from various vertebrate groups would be important to better understand labyrinth sizefunction relationships. Melville Jones & Spells (1963) have shown that fishes have large labyrinth sizes compared to most tetrapods, but they don't have lungfish data and the large labyrinth sizes of fishes have often remained uncommented on in tetrapod works. I think this study offers a fantastic opportunity to provide comparative labyrinth size data for lungfishes. In this regard, it would be really interesting to quantify labyrinth size relative to head size, and show a respective (phylogenetic) regression analysis. Ideally, the size of the labyrinth could be quantified along the arc lengths of the semicircular canals, but other ways are also thinkable (for example a box volume of labyrinth size by the existing measurements, contrasted with a box volume of the skull, i.e. heightwidthlength).

    Firstly, many thanks for the suggested reading of Bronzati et al. (2021) And while we consider a labyrinth skull size regression analysis to be a worthwhile suggestion, we have chosen not to include one in this study, partly as there is no phylogenetic regression based on the new methods that we are using, and secondly that it forms the basis of another study currently underway by some of the authors.

  2. Evaluation Summary:

    Clement and colleagues describe and illustrate the endocasts of six Palaeozoic lungfish genera from superb 3D fossil material, which are very informative for the understanding of brain evolution of lungfishes, the extant sister group to land vertebrates. Rendering important anatomical details regarding brain evolution in lungfishes, and sarcopterygians in general, this work will be of broad interest to zoologists, including vertebrate paleontologists and neuroanatomists.

    (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 #, Reviewer #2 and Reviewer #3 agreed to share their name with the authors.)

  3. Reviewer #1 (Public Review):

    On the basis of 13 endocast data including six new lungfish data, Clement and colleagues conducted the multivariate morphometric analyses using 17 neurocranial variables, and revealed quantitatively several transformational trends of brain in lungfish evolution. With its important anatomical details, this work will be of broad interest to many scientists. The inference in the manuscript is overall clear, and the conclusions are well supported by data. Re-organization of figures and changes to the figures and the text will improve the readability of the manuscript.

  4. Reviewer #2 (Public Review):

    First, I want to congratulate the author team on this manuscript, which I read with great pleasure. I think this will be a fine addition to the literature!

    The present MS by Clement et al. provides a comprehensive overview of the brain shapes of lungfishes. Besides previously known/described brain endocasts, the work includes models and descriptions of previously undescribed taxa. Notably, all CT data are deposited online following best practices when working with digital anatomy. The specimen sample is impressive, especially as the sampled material is housed in museum all over the world. Although the sample size may seem numerically low (12 taxa), this actually is a comprehensive sample of fossil (and extant) lungfishes in terms of what's preserved in the first place.

    The study at hand has several goals: (1) The description of lungfish brains for taxa that were previously undescribed; (2) the quantification of aspects of brain shape using morphometric measurements; (3) the characterization of brain shape evolution of lungfishes using exploratory methods that ordinate morphometric measurements into a morphospace.

    The provided 3D data and descriptions will serve as valuable comparisons in future lungfish work. This type of data is imperial for palaeontological studies in general, and the anatomical information will be extremely valuable in the future. For example, anatomical characters related to brain architecture have been shown to be informative about phylogeny in the past, and the presented data may inform future phylogenetic studies.

    The quantification of brain shape via (largely linear) measurements is relatively simplistic, and can thus only detect gross trends in brain shape evolution among lungfishes. The authors describe several such trends - such as high variation in the olfactory brain region in comparison to other parts of the brain. The results and interpretations drawn from the authors are supported by their data, and the approach taken is valid, even if more sophisticated shape quantification methods (e.g. 3D landmarking) and analytical methods (e.g. explicit phylogenetic comparative methods) are available, which could provide additional insights in the future. The presented results and interpretations in this regard must be seen as a preliminary assessment of lungfish brain evolution, but it is clearly written and generally well performed.

    A potential shortcoming of the paper is the lack of explicit hypothesis testing, which is not problematic per se, but puts limits on the conclusions the authors can draw from their data. For example, the authors state that different anatomical parts of the labyrinth (particularly, the utricle with respect to the semicircular canals or saccule) may show modular dissociation from other labyrinth modules, based on the polarity of eigenvalue signs of the PCA analysis. I think this is fine as a first approximation, but of course there are explicit statistical tools available to test for modularity/integration, such as two-block partial least squares regression analysis (Rohlf & Corti 2000, Syst. Biol.). I don't see the lack of usage of such methods as problematic, because you cannot do everything in one paper, and the authors remain careful in their interpretation. It may be advisable, however, to add the odd sentence or statement about how some findings are preliminary or hypothesized, and that these should receive further treatment and testing using other methods in the future. I think this approach is actually very rewarding, because then you can inspire future work by outlining outstanding research problems that arise from the new data presented herein.

    In the following, I comment on a few aspects of the manuscripts. These represent instances where I had additional thoughts or ideas on how to slightly improve various aspects of the manuscript.

    1. Presentation of PCA results

    The authors provide several PCA analyses (preliminary analyses on partial matrices, BPCA, InDaPCA), and are very explicit about the procedures in general. For instance, I appreciate they explicitely state using correlation matrices for PCA analyses due to the usage of different measurement units among their data.

    Visually, the BPCA and InDaPCA are presented in figures 2 and 3, whereas the preliminary partial matrix PCAs are only reported as supplementary figures. While I don't object to any of this, I find the sequence of information given in the results section suboptimal.

    The authors start by discussing the partial matrix analyses, although none of these analyses are visually/graphically depicted in the main text figures, and although their results do not seem to be of real importance for the narrative of the discussion. The other two PCA analyses actually are presented afterwards and separately, but they convey some common signals, particularly that the major source of variation seems to be a decreasing olfactory angle with increasing olfactory length, and a scaling relationship between all linear measurements (which all have the same eigenvector signs on the first PC axis). I wonder if an alternative way of presenting the PCA results would be better for this particular MS. For example, the authors could give "first level observations" first ("PCA analyses agree in X,Y,Y"), and then move to second order observations ("Morphospace of BPCA has some interesting taxon distribution with regard to chirodipterids"; "InDaPCA axis projections continuously retrieve clustering of specific variables"). I suspect this would shorten the text somewhat and could serve as a clearer articulation of the take home messages?

    2. Selection of PC axes for interpretation

    You describe how you use the broken-stick method to decide how many PC axes are retained for the interpretation of results, which I agree is a good procedure. However, I have a few questions regarding this.

    First, in line 331 (description of InDaPCA) you state that the first three axes are non-trivial "based on the screeplot" - which got me confused because it sounds a bit like eyeballing off the screeplot. Have you used the broken stick method for all your PCA analyses?

    The second question relates to the results of the broken stick method, which I did not find reported. Unless I am mistaken, for the xth axis, the method sums the fractions of 1/i (whereby i = x..n; n = number of axes), and divides this number by n to get a value of expected variation per axis. This number is then compared with the actual value of variance explained by the axis. So for the 1st of 17 axes, the broken-stick expectation is = (1 + 1/2 + .. + 1/17) / 17. If you apply this to your BPCA, the third axis' value (i.e., (1/3 + ... + 1/17)/17) is 0.114, which is smaller than the reported 0.120 that PC3 explains. Thus, following the broken stick method, PC3 does explain more variation that expected (and should thus be retained, contra your comment in line 311 which refers to two non-trivial axes)? Related to this potential issue is the presentation of the BPCA results in Fig. 2: You present loadings of three PC axes, although only the first two are considered in morphospace bi-plots and although the text also mentions only two non-trival axes. If the third axis is indeed non-trivial, then the loading-presentation could be retained in the figure, but then the authors should consider showing a PC1 vs. PC3 plot in addition to the currently presented biplot showing the first and second axis only. If the third axis indeed is trivial, as currently suggested by the text, then showing the loadings is unnecessary.

    It would be great if you clarify the usage/application of the broken stick method for all your PCAs. An easy way to report the results may be the add a row to each of your PCA loading tables in the supplements, in which you divide the actual value of variation explained by the value expected under the broken stick method - this way, all axes which explain more variation than expected by the stick method have values larger than 1, and axes which explain less have values lower than 1.

    3. Missing commentary on allometry

    In basically all PCA analyses, the first PC axis seems to be dominated by allometric size effects, given that all linear measurements have the same eigenvalue signs. The authors do acknowledge this (lines 314-316; 335-336), but offer no further comment on size effects/allometry. For example, it would be interesting to see how the linear measurements scale with overall head size. Similarly, the authors note that the semicircular canal measurements covary strongly, as do the utricle and saccule height/length measurements (paragraph line 346). Basically, it seems that the semicircular canal measurements scale with one another: as one gets bigger, so gets the other. It is interesting that the utricle does not seem to follow the same scaling pattern as the saccule and semicircular canals, and it would be good to hear if the authors think that there is a functional implication for this. Increases in utricular/saccular/semicircular canal sizes are usually explained by increased sensitivity - so is an increased utricular size a compensatory development to decreased semicircular canal+saccule size to retain an overall level of sensitivity, or does it maybe related to a relative change of importance of the specific functions, e.g. increased importance of linear accelerations in the horizontal plane with simultaneous decrease of importance of angular and vertical accelerations?

    4. Labyrinth size

    With the above mentioned utricular exception, labyrinth size measurements particularly on the semicircular canals seem to imply that there is a relative consistent scaling relationship between the canals. When one canal gets larger, so do the others, perhaps thereby retaining canal symmetry across different absolute labyrinth sizes. Labyrinth size in tetrapods is often interpreted in relation to body size/mass or head size (e.g. Melville Jones & Spells 1963, Proc. R. Soc. Lond. Biol. Sci.; Spoor & Zonneveldt 1998, Yearb. Phys. Anthr.; Spoor et al. 2002, Nature; Spoor et al. 2007, PNAS; Bronzati et al. 2021, Curr. Biol.), as deviations from the expected labyrinth size per head size indicate increased or decreased relative labyrinth sensitivities. Large relative head sizes of birds and (within) mammals have generally been interpreted as indicative of "active" or "agile" behaviour, although doubt has been casted on these relationships recently (e.g., Bronzati et al. 2021). Increased sampling of relative labyrinth size from various vertebrate groups would be important to better understand labyrinth size-function relationships. Melville Jones & Spells (1963) have shown that fishes have large labyrinth sizes compared to most tetrapods, but they don't have lungfish data and the large labyrinth sizes of fishes have often remained uncommented on in tetrapod works. I think this study offers a fantastic opportunity to provide comparative labyrinth size data for lungfishes. In this regard, it would be really interesting to quantify labyrinth size relative to head size, and show a respective (phylogenetic) regression analysis. Ideally, the size of the labyrinth could be quantified along the arc lengths of the semicircular canals, but other ways are also thinkable (for example a box volume of labyrinth size by the existing measurements, contrasted with a box volume of the skull, i.e. height*width*length).

  5. Reviewer #3 (Public Review):

    This manuscript deals with the anatomical diversity of Paleozoic lungfishes through a descriptive and quantitative lens that makes possible to draw comparisons between taxa and make tentative correlations between form and function on lungfish endocasts. The work redescribes the morphology of six lungifish endocasts based on virtual reconstructions obtained through CT-scanning. Then, the authors put these virtual models into a quantitative context through principal component analysis which highlights interesting morphological features and the morphological diversity across Devonian and extant forms.

    The anatomical description of the endocasts is well-detailed and add to the knowledge of lungfish neuroanatomy. Some of these endocasts have been previously described in the literature but the authors re-description adds considerably to the original descriptions. Additionally, by having these descriptions together in a context they provide a good starting point for future anatomical comparison of the studied taxa.

    The quantitative analysis of the endocast models was conducted through principal component analysis (PCA) and other iterations of this methodology (BPCA and InDaPCA) of selected meristics of these endocasts. This analysis demonstrates that there is clustering of some Devonian forms while others show more extreme morphologies. The meristics used might not represent the complex morphology of these endocasts on its totality, but a reductive approach as this is sound with an exploratory analysis as the one presented here. However, the most interesting results are those related to the relative contribution of different structures for the observed variation and proposed interactions between different regions (e.g. relation between sacculus - utriculus and semicircular canals). The authors clearly demonstrate that the forebrain and olfactory capsules are highly plastic while the rhombencephalon seems to be more conservative, to the exception of the inner ear.