Cross-species alignment along the chronological axis reveals evolutionary effect on structural development of the human brain

  1. Yue Li
  2. Qinyao Sun
  3. Shunli Zhu
  4. Congying Chu
  5. Jiaojian Wang

  • Curated by eLife

    eLife logo

    eLife assessment

    This important study compared the brain development trajectories of humans and macaque monkeys to quantify different evolutionary effects of convergent and divergent neural pathways between the two species. The cross-species evidence is solid, based on brain age prediction models that were carefully developed by using public MRI datasets of both humans and macaque monkeys. The findings will be of interest to neuroscientists, developmental biologists, and evolutionary biologists.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Disentangling the evolution mysteries of the human brain has always been an imperative endeavor in neuroscience. Although many previous comparative studies revealed genetic, brain structural and connectivity distinctness between human and other nonhuman primates, the brain evolutional mechanism is still largely unclear. Here, we proposed to embed the brain anatomy of human and macaque in the developmental chronological axis to construct cross-species predictive model to quantitatively characterize brain evolution using two large public human and macaque datasets. We observed that applying the trained models within-species could well predict the chronological age. Interestingly, we found the model trained in macaque showed a higher accuracy in predicting the chronological age of human than the model trained in human in predicting the chronological age of macaque. The cross-application of the trained model introduced an individual brain cross-species age gap index to quantify the cross-species discrepancy along the temporal axis of brain development and was found to be associated with the behavioral performance in visual acuity test and picture vocabulary test in human. Taken together, our study situated the cross-species brain development along the chronological axis, which highlighted the disproportionately anatomical development in human brain to extend our understanding of the potential evolutionary effects.

Article activity feed

  1. Version published to 10.7554/elife.96020 on eLife
  2. eLife

    Author response:

    Reviewer #1 (Public Review):

    The authors conducted cross-species comparisons between the human brain and the macaque brain to disentangle the specific characteristics of structural development of the human brain. Although previous studies had revealed similarities and differences in brain anatomy between the two species by spatially aligning the brains, the authors made the comparison along the chronological axis by establishing models for predicting the chronological ages with the inputting brain structural features. The rationale is actually clear given that brain development occurs over time in both. More interestingly, the model trained on macaque data was better able to predict the age of humans than the human-trained model was at predicting macaque age. This revealed a brain cross-species age gap (BCAP) that quantified the discrepancy in brain development between the two species, and the authors even found this BCAP measure was associated with performance on behavioral tests in humans. Overall, this study provides important and novel insights into the unique characteristics of human brain development. The authors have employed a rigorous scientific approach, reflecting diligent efforts to scrutinize the patterns of brain age models across species. The clarity of the rationale, the interpretability of the methods, and the quality of the presentation all contribute to the strength of this work.

    We are grateful to your helpful and thorough review and for being so positive about our manuscript. Following your recommendations, we have added more analytic details that have strengthened our paper. We would like to thank you for your input.

    Reviewer #2 (Public Review):

    In the current study, Li et al. developed a novel approach that aligns chronological age to a cross-species brain age prediction model to investigate the evolutionary effect. This method revealed some interesting findings, like the brain-age gap of the macaque model in predicting human age will increase as chronological age increases, suggesting an evolutionary alignment between the macaque brain and the human brain in the early stage of development. This study exhibits ample novelty and research significance. However, I still have some concerns regarding the reliability of the current findings.

    We thank you for the positive and appreciative feedback on our work and the insightful comments, which we have addressed below.

    Question 1: Although the authors named their new method a "cross-species" model, the current study only focused on the prediction between humans and macaques. It would be better to discuss whether their method can also generalize to cross-species examination of other species (e.g., C. elegans), which may provide more comprehensive evolutionary insights. Also, other future directions with their new method are worth discussing.

    We appreciate your insightful comment regarding the generalizability of our model to other species. As you said, we indeed only performed human-macaque cross-species study not including other species. In our study, we only focused human and macaque because macaque is considered to be one of the closest primates to humans except chimpanzees and thus is considered to be the best model for studying human brain evolution. However, our proposed method has limitations that limit its generalizability for other species, e.g., C. elegans. First, our model was trained using MRI data, which limits its applicability to species for which such data is unavailable. This technological requirement brings a barrier to broaden cross-species application. Second, our current model is based on homologous brain atlases that are available for both humans and macaques. The lack of comparable atlases for other species further restricts the model's generalizability. We have discussed this limitation in the revised manuscript and outlined potential future directions to overcome these challenges. This includes discussing the need for developing comparable imaging techniques and standardized brain atlases across a wider range of species to enhance the model's applicability and broaden our understanding of cross-species neurodevelopmental patterns.

    On page 15, lines 11-18

    “However, the existing limitation should be noted regarding the generalizability of our proposed approach for cross-species brain comparison. Our current model relies on homologous brain atlases, and the lack of comparable atlases for other species restricts its broader applicability. To address this limitation, future research should focus on developing prediction models that do not depend on atlases. For instance, 3D convolutional neural networks could be trained directly on raw MRI data for age prediction. These deep learning models may offer greater flexibility for cross-species applications once the training within species is complete. Such advancements would significantly enhance the model's adaptability and expand its potential for comparative neuroscience studies across a wider range of species.”

    Question 2: Algorithm of prediction model. In the method section, the authors only described how they chose features, but did no description about the algorithm (e.g., supporting vector regression) they used. Please add relevant descriptions to the methods.

    Thank you for your comment. We apologize for not providing sufficient details about the model training process in our initial submission. In our study, we used a linear regression model for prediction. We have provided more details regarding the algorithm of prediction model in our response to Reviewer #1. For your convenience, we have attached them below.

    For details on the algorithm of prediction model:

    “A linear regression model was adopted for intra- and inter-species age prediction. The linear regression model was built including the following three main steps: 1) Feature selection: a total of two steps are required to extract the final features. The first step is preliminary extraction. First, all the human or macaque participants were divided into 10-fold and 9-fold was used for model training and 1-fold for model test. The preliminary features were chosen by identifying the significantly age-associated features with p < 0.01 during calculating Pearson’s correlation coefficients between all the 260 features and actual ages of the 9-fold subjects. This process was repeated 100 times. Since we obtained not exactly the same preliminary features each time, we thus further analyzed the preliminary features using two methods to determine the final features: common features and minimum mean absolute error (min MAE). Common features are the preliminary features that were selected in all the 100 times during preliminary model training. The min MAE features were the preliminary features that with the smallest MAE value during the 100 times model test for predicting age. After the above feature selections, we obtained two sets of features: 62 macaque features and 225 human features (common features) and 117 macaque features and 239 human features (min MAE). In addition, to further exclude the influences of unequal number of features in human and macaque, we also selected the first 62 features in human and macaque to test the model prediction performances. 2) Model construction: we conducted age prediction linear model using 10-fold cross-validation based on the selected features for human and macaque separately. The linear model parameters are obtained using the training set data and applied to the test set for prediction. The above process is also repeated 100 times. 3) Prediction: with the above results, we obtained the optimal linear prediction models for human and macaque. Next, we performed intra-species and inter-species brain age prediction, i.e., human model predicted human age, human model predicted macaque age, macaque model predicted macaque age and macaque model predicted human age. Three sets of features (62 macaque features and 225 human features; 117 macaque features and 239 human features; 62 macaque features and 62 human features) were used to test the prediction models for cross-validation and to exclude effects of different number of features in human and macaque. In the main text, we showed the results of brain age prediction, brain developmental and evolutional analyses based on common features and the results obtained using other two types of features were shown in supplementary materials. The prediction performances were evaluated by calculating the Pearson’s correlation and MAE between actual ages and predicted ages.”

    Question 3: Sex difference. The sex difference results are strange to me. For example, in the second row of Figure Supplement 3A, different models show different correlation patterns, but why their Pearson's r is all equal to 0.3939? If they are only typo errors, please correct them. The authors claimed that they found no sex difference. However, the results in Figure Supplement 3 show that, the female seems to have poorer performance in predicting macaque age from the human model. Moreover, accumulated studies have reported sex differences in developing brains (Hines, 2011; Kurth et al., 2021). I think it is also worth discussing why sex differences can't be found in the evolutionary effect.

    Reference:

    Hines, M. (2011). Gender development and the human brain. Annual review of neuroscience, 34, 69-88.

    Kurth, F., Gaser, C., & Luders, E. (2021). Development of sex differences in the human brain. Cognitive Neuroscience, 12(3-4), 155-162.

    It is recommended that the authors explore different prediction models for different species. Maybe macaques are suitable for linear prediction models, and humans are suitable for nonlinear prediction models.

    Thank you for pointing the typos out and comments on sex difference. In Figure Supplement 3A, there are typos for Pearson’s r values and we have corrected it in updated Figure 2-figure supplement 3. For details, please see the updated Figure 2-figure supplement 3 and the following figure.

    Regarding gender effects, we acknowledge your point about the importance of gender differences in understanding brain evolution and development. In our study, however, our primary goal was to develop a robust age prediction model by maximizing the number of training samples. To mitigate gender-related effects in our main results, we incorporated gender information as a covariate in the ComBat harmonization process. We conducted a supplementary analysis just to demonstrate the stability of our proposed cross-species age prediction model by separating the data with gender variable not to investigate gender differences. Although our results demonstrated that gender-specific models could still significantly predict chronological age, we refrained from emphasizing these models' performance in gender-specific species comparisons due to difficulty in explanation for the predicted gender difference. For cross-species prediction, whether a higher Pearson’s r value between actual age and predicted age could reflect conserved evolution for male or female is not convincing. In addition, we adopted same not different prediction models for human and macaque aiming to establish a comparable model between species. Generally speaking, the nonlinear model could obtain better prediction accuracy than linear model. If different species used different models, it is unfair to perform cross-species prediction. Importantly, our study aimed to developed new index based on the same prediction models to quantify brain evolution difference, i.e., brain cross-species age gap (BCAP) instead of traditional statistical analyses. Different prediction models for different species may introduce bias causing by prediction methods and thus impacting the accuracy of BCAP. Thus, we adopted the linear model with best prediction performances for intra-species prediction in this study for cross-species prediction. Although our main goal in this study is to set up stable cross-species prediction model and the models built using either male or female subjects showed good performances during cross-species prediction, however, as your comment, how to unbiasedly characterize evolutionary gender differences using machining learning approaches needs to be further investigated since there are many reports about the gender difference in developing brain in humans. In fact, whether macaque brains have the same gender differences as humans is an interesting scientific question worth studying. Thus, we have included a discussion on how to use machining learning method to study the evolutionary gender difference in our revised manuscript.

    On page 15, lines 18-23 and page 16, line 1-4

    “Many studies have reported sex differences in developing human brains (Hines, 2011; Kurth, Gaser, & Luders, 2021), however, whether macaque brains have similar sex differences as humans is still unknown. We used machining learning method for cross-species prediction to quantify brain evolution and the established prediction models are stable even when only using male or female data, which may indicate that the proposed cross-species prediction model has no evolutionary sex difference. Although the stable prediction model can be established in either male or female participants for cross-species prediction, this indeed does not mean that there are no evolutionary sex differences due to lack of quantitative comparative analysis. In the future, we need to develop more objective, quantifiable and stable index for studying sex differences using machining learning methods to further identify sex differences in the evolved brain”

    Reviewer #3 (Public Review):

    The authors identified a series of WM and GM features that correlated with age in human and macaque structural imaging data. The data was gathered from the HCP and WA studies, which was parcellated in order to yield a set of features. Features that correlated with age were used to train predictive intra and inter-species models of human and macaque age. Interestingly, while each model accurately predicted the corresponding species age, using the macaque model to predict human age was more accurate than the inverse (using the human model to predict macaque age). In addition, the prediction error of the macaque model in predicting human age increased with age, whereas the prediction error of the human model predicting macaque age decreased with age.

    After elaboration of the predictive models, the authors classified the features for prediction into human-specific, macaque-specific and common to human and macaque, where they most notably found that macaque-only and common human-macaque areas were located mainly in gray matter, with only a few human-specific features found in gray matter. Furthermore, the authors found significant correlations between BCAP and picture vocabulary (positive correlation) test and visual sensitivity (negative correlation) test. Several white matter tracts (AF, OR, SLFII) were also identified showing a correlation with BCAP.

    Thank you for providing this excellent summary. We appreciate your thorough review and concise overview of our work.

    STRENGTHS AND WEAKNESSES

    The paper brings an interesting perspective on the evolutionary trajectories of human and non-human primate brain structure, and its relation to behavior and cognition. Overall, the methods are robust and support the theoretical background of the paper. However, the overall clarity of the paper could be improved. There are many convoluted sentences and there seems to be both repetition across the different sections and unclear or missing information. For example, the Introduction does not clearly state the research questions, rather just briefly mentions research gaps existing in the literature and follows by describing the experimental method. It would be desirable to clearly state the theoretical background and research questions and leave out details on methodology. In addition, the results section repeats a lot of what is already stated in the methods. This could be further simplified and make the paper much easier to read.

    In the discussion, authors mention that "findings about cortex expansion are inconsistent and even contradictory", a more convincing argument could be made by elaborating on why the cortex expansion index is inadequate and how BCAP is more accurate.

    Thank you for highlighting the interesting aspects of our work. We are sorry for the lack of the clarity in certain parts of our manuscript. Following your valuable suggestions, we have revised the manuscript to reduce unnecessary repetitions and provide a clearer statement of our research question in Introduction. Specifically, unlike previous analyses of human and macaque evolution using comparative neuroscience, this study embeds chronological axis into the cross-species evolutionary analysis process. It constructed a linear prediction model of brain age for humans and macaques, and quantitatively described the degree of evolution. The brain structure based cross-species age prediction model and cross-species brain age differences proposed in this study further eliminate the inherent developmental effects of humans and macaques on cross-species evolutionary comparisons, providing new perspectives and approaches for studying cross-species development. Regarding the existing repetition in the results section, we have simplified them for the clarity. Regarding the comparison between the cortex expansion index and BCAP, we would like to emphasize that the cortex expansion index was derived without fully considering cross-species alignment along the chronological axis. Specifically, this index does not correspond to a specific developmental stage, but rather focuses on a direct comparison between the two species. In contrast, BCAP addresses this limitation by utilizing a prediction model to establish alignment (or misalignment) between species at the individual level. Therefore, BCAP may serve as a more flexible and nuanced tool for cross-species brain comparison.

    STUDY AIMS AND STRENGTH OF CONCLUSIONS

    Overall, the methods are robust and support the theoretical background of the paper, but it would be good to state the specific research questions -even if exploratory in nature- more specifically. Nevertheless, the results provide support for the research aims.

    Thank you for excellent suggestion. We have revised our introduction to state the specific research question as mentioned above.

    IMPACT OF THE WORK AND UTILITY OF METHODS AND DATA TO THE COMMUNITY

    This study is a good first step in providing a new insight into the neurodevelopmental trajectories of humans and non-human primates besides the existing cortical expansion theories.

    Thank you for your encouraging comment.

    ADDITIONAL CONTEXT:

    It should be clearly stated both in the abstract and methods that the data used for the experiment came from public databases.

    Thank you for your suggestion. We have added this information in both abstract and method. For details, please see page 2, line 9 in Abstract section; page 16, lines 10-11 and page 17, lines 6-10 in Materials and Method section.

  3. eLife

    eLife assessment

    This important study compared the brain development trajectories of humans and macaque monkeys to quantify different evolutionary effects of convergent and divergent neural pathways between the two species. The cross-species evidence is solid, based on brain age prediction models that were carefully developed by using public MRI datasets of both humans and macaque monkeys. The findings will be of interest to neuroscientists, developmental biologists, and evolutionary biologists.

  4. eLife

    Reviewer #1 (Public Review):

    The authors conducted cross-species comparisons between the human brain and the macaque brain to disentangle the specific characteristics of structural development of the human brain. Although previous studies had revealed similarities and differences in brain anatomy between the two species by spatially aligning the brains, the authors made the comparison along the chronological axis by establishing models for predicting the chronological ages with the inputting brain structural features. The rationale is actually clear given that brain development occurs over time in both. More interestingly, the model trained on macaque data was better able to predict the age of humans than the human-trained model was at predicting macaque age. This revealed a brain cross-species age gap (BCAP) that quantified the discrepancy in brain development between the two species, and the authors even found this BCAP measure was associated with performance on behavioral tests in humans. Overall, this study provides important and novel insights into the unique characteristics of human brain development. The authors have employed a rigorous scientific approach, reflecting diligent efforts to scrutinize the patterns of brain age models across species. The clarity of the rationale, the interpretability of the methods, and the quality of the presentation all contribute to the strength of this work.

  5. eLife

    Reviewer #2 (Public Review):

    In the current study, Li et al. developed a novel approach that aligns chronological age to a cross-species brain age prediction model to investigate the evolutionary effect. This method revealed some interesting findings, like the brain-age gap of the macaque model in predicting human age will increase as chronological age increases, suggesting an evolutionary alignment between the macaque brain and the human brain in the early stage of development. This study exhibits ample novelty and research significance. However, I still have some concerns regarding the reliability of the current findings.

    (1) Although the authors named their new method a "cross-species" model, the current study only focused on the prediction between humans and macaques. It would be better to discuss whether their method can also generalize to cross-species examination of other species (e.g., C. elegans), which may provide more comprehensive evolutionary insights. Also, other future directions with their new method are worth discussing.

    (2) Algorithm of prediction model. In the method section, the authors only described how they chose features, but did no description about the algorithm (e.g., supporting vector regression) they used. Please add relevant descriptions to the methods.

    (3) Sex difference. The sex difference results are strange to me. For example, in the second row of Figure Supplement 3A, different models show different correlation patterns, but why their Pearson's r is all equal to 0.3939? If they are only typo errors, please correct them. The authors claimed that they found no sex difference. However, the results in Figure Supplement 3 show that, the female seems to have poorer performance in predicting macaque age from the human model. Moreover, accumulated studies have reported sex differences in developing brains (Hines, 2011; Kurth et al., 2021). I think it is also worth discussing why sex differences can't be found in the evolutionary effect.

    Reference:
    Hines, M. (2011). Gender development and the human brain. Annual review of neuroscience, 34, 69-88.
    Kurth, F., Gaser, C., & Luders, E. (2021). Development of sex differences in the human brain. Cognitive Neuroscience, 12(3-4), 155-162.

  6. eLife

    Reviewer #3 (Public Review):

    SUMMARY:

    The authors identified a series of WM and GM features that correlated with age in human and macaque structural imaging data. The data was gathered from the HCP and WA studies, which was parcellated in order to yield a set of features. Features that correlated with age were used to train predictive intra and inter-species models of human and macaque age. Interestingly, while each model accurately predicted the corresponding species age, using the macaque model to predict human age was more accurate than the inverse (using the human model to predict macaque age). In addition, the prediction error of the macaque model in predicting human age increased with age, whereas the prediction error of the human model predicting macaque age decreased with age.

    After elaboration of the predictive models, the authors classified the features for prediction into human-specific, macaque-specific and common to human and macaque, where they most notably found that macaque-only and common human-macaque areas were located mainly in gray matter, with only a few human-specific features found in gray matter. Furthermore, the authors found significant correlations between BCAP and picture vocabulary (positive correlation) test and visual sensitivity (negative correlation) test. Several white matter tracts (AF, OR, SLFII) were also identified showing a correlation with BCAP.

    STRENGTHS AND WEAKNESSES

    The paper brings an interesting perspective on the evolutionary trajectories of human and non-human primate brain structure, and its relation to behavior and cognition. Overall, the methods are robust and support the theoretical background of the paper. However, the overall clarity of the paper could be improved. There are many convoluted sentences and there seems to be both repetition across the different sections and unclear or missing information. For example, the Introduction does not clearly state the research questions, rather just briefly mentions research gaps existing in the literature and follows by describing the experimental method. It would be desirable to clearly state the theoretical background and research questions and leave out details on methodology.
    In addition, the results section repeats a lot of what is already stated in the methods. This could be further simplified and make the paper much easier to read.
    In the discussion, authors mention that "findings about cortex expansion are inconsistent and even contradictory", a more convincing argument could be made by elaborating on why the cortex expansion index is inadequate and how BCAP is more accurate.

    STUDY AIMS AND STRENGTH OF CONCLUSIONS

    Overall, the methods are robust and support the theoretical background of the paper, but it would be good to state the specific research questions -even if exploratory in nature- more specifically. Nevertheless, the results provide support for the research aims.

    IMPACT OF THE WORK AND UTILITY OF METHODS AND DATA TO THE COMMUNITY

    This study is a good first step in providing a new insight into the neurodevelopmental trajectories of humans and non-human primates besides the existing cortical expansion theories.

    ADDITIONAL CONTEXT:

    It should be clearly stated both in the abstract and methods that the data used for the experiment came from public databases.

  7. Version published to 10.1101/2024.02.27.582251v1 on bioRxiv