Pan-tissue Transcriptome Analysis Reveals Sex-dimorphic Human Aging

Reviewer #1 (Public review):

Summary:

Wang et al. investigate sexual dimorphic changes in the transcriptome of aged humans. This study relies upon analysis of the Genotype-Tissue Expression dataset that includes 54 tissues from human donors. The authors investigate 17,000 transcriptomes from 35 tissues to investigate the effect of age and sex on transcriptomic variation, including the analysis of alternative splicing. Alternative splicing is becoming more appreciated as an influence in the aging process, but how it is affected by sexual dimorphism is still largely unclear. The authors investigated multiple tissues but ended up distilling brain tissue down to four separate regions: decision, hormone, memory, and movement. Building upon prior work, the authors used an analysis method called principal component-based signal-to-variation ratio (pcSVR) to quantify differences between sex or age by considering data dispersion. This method also considers differentially expressed genes and alternative splicing events.

Strengths:

(1) The authors investigate sexual dimorphism on gene expression and alternative splicing events with age in multiple tissues from a large publicly available data set that allows for reanalysis.

(2) Furthermore, the authors take into account the ethnic background of donors. Identification of aging-modulating genes could be useful for the reanalysis of prior data sets.

Weaknesses:

The models built off of the GTEx dataset should be tested in another data set (ex. Alzheimer's disease) where there are functional changes that can be correlated. Gene-length-dependent transcription decline, which occurs with age and disease, should also be investigated in this data set for potential sexual dimorphism.

Reviewer #2 (Public review):

Summary:

In this manuscript, Wang et al analyze ~17,000 transcriptomes from 35 human tissues from the GTEx database and address transcriptomic variations due to age and sex. They identified both gene expression changes as well as alternative splicing events that differ among sexes. Using breakpoint analysis, the authors find sex dimorphic shifts begin with declining sex hormone levels with males being affected more than females. This is an important pan-tissue transcriptomic study exploring age and sex-dependent changes although not the first one.

Strengths:

(1) The authors use sophisticated modeling and statistics for differential, correlational, and predictive analysis.

(2) The authors consider important variables such as genetic background, ethnicity, sampling bias, sample sizes, detected genes, etc.

(3) This is likely the first study to evaluate alternative splicing changes with age and sex at a pan-tissue scale.

(4) Sex dimorphism with age is an important topic and is thoroughly analyzed in this study.

Weaknesses:

(1) The findings have not been independently validated in a separate cohort or through experiments. Only selective splicing factor regulation has been verified in other studies.

(2) It seems the authors have not considered PMI or manner of death as a variable in their analysis.

(3) The manuscript is very dense and sometimes difficult to follow due to many different types of analyses and correlations.

(4) Short-read data can detect and quantify alternative splicing events with only moderate confidence and therefore the generalizability of these findings remains to be experimentally validated.

Reviewer #3 (Public review):

Summary:

In this study, Wang et al utilized the available GTEx data to compile a comprehensive analysis that attempt to reveal aging-related sex-dimorphic gene expression as well as alternative splicing changes in humans.

The key conclusions based on their analysis are that

(1) extensive sex-dimorphisms during aging with distinct patterns of change in gene expression and alternative splicing (AS), and

(2) the male-biased age-associated AS events have a stronger association with Alzheimer's disease, and

(3) the female-biased events are often regulated by several sex-biased splicing factors that may be controlled by estrogen receptors. They further performed break-point analysis and revealed that in males there are two main breakpoints around ages 35 and 50, while in females, there is only one breakpoint at 45.

Strengths:

This study sets an ambitious goal, leveraging the extensive GTEx dataset to investigate aging-related, sex-dimorphic gene expression and alternative splicing changes in humans. The research addresses a significant question, as our understanding of sex-dimorphic gene expression in the context of human aging is still in its early stages. Advancing our knowledge of these molecular changes is vital for identifying therapeutic targets for age-related diseases and extending the human health span. The study is highly comprehensive, and the authors are commendable for their attempted thorough analysis of both gene expression and alternative splicing - an area often overlooked in similar studies.

Weaknesses:

Due to the inherent noise within the GTEx dataset - which includes numerous variables beyond aging and sex - there are significant technical concerns surrounding this study. Additionally, the lack of cross-validation with independent, existing data raises questions about whether the observed gene expression changes genuinely reflect those associated with human aging. For instance, the break-point analysis in this study identifies two major breakpoints in males around ages 35 and 50, and one breakpoint in females at age 45; however, these findings contradict a recent multi-omics longitudinal study involving 108 participants aged 25 to 75 years, where breakpoint at 44 and 60 years was observed in both male and females (Shen et al, 2024). These issues cast doubt on the robustness of the study's conclusions. Specific concerns are outlined below:

(1) The primary method used in this study is linear regression, incorporating age, sex, and age-by-sex interactions as covariates, alongside other confounding factors (such as ethnicity) as unknown variables. However, the analysis overlooks two critical known variables in the GTEx dataset: time of death (TOD) and postmortem interval (PMI). Both TOD and PMI are recorded for each sample and account for substantial variance in gene expression profiles. A recent study by Wucher et al.(Wucher et al, 2023) demonstrated the powerful impact of TOD on gene expression by using it to reconstruct human circadian and even circannual datasets. Similarly, Ferreira et al. (Ferreira et al, 2018) highlighted PMI's influence on gene expression patterns. Without properly adjusting for these two variables, confidence in the study's conclusions remains limited at best.

(2) To demonstrate that their analysis is robust and that the covariates TOD and PMI are otherwise negligible - the authors should cross-validate their findings with independent datasets to confirm that the identified gene expression changes are reproducible for some tissues. For instance, the recent study by Shen et al. (Shen et al., 2024) in Nature Aging offers an excellent dataset for cross-validation, particularly for blood samples. Comparing the GTEx-derived results with this longitudinal transcriptome dataset would enable verification of gene expression changes at both the individual gene and pathway levels. Without such validation, confidence in the study's conclusions remains limited.

(3) As a demonstration of the lack of such validation, in the Shen et al. study (Shen et al., 2024), breakpoints at 44 and 60 years were observed in both males and females, while this study identifies two major breakpoints in males around ages 35 and 50, and one breakpoint in females at age 45. What caused this discrepancy?

(4) Although the alternative splicing analysis is intriguing, the authors did not differentiate between splicing events that alter the protein-coding sequence and those that do not. Many splicing changes occurring in the 5' UTR and 3' UTR regions do not impact protein coding, so it is essential to filter these out and focus specifically on alternative splicing events that can modify protein-coding sequences.

(5) One of the study's main conclusions - that "male-biased age-associated AS events have a stronger association with Alzheimer's disease" - is not supported by the data presented in Figure 4A, which shows an association with "regulation of amyloid precursor formation" only in female, not male, alternative splicing genes. Additionally, the gene ontology term "Alzheimer's disease" is absent from the unbiased GO analysis in Figure S6. These discrepancies suggest that the focus on Alzheimer's disease may reflect selective data interpretation rather than results driven by an unbiased analysis.

(6) The experimental data presented in Figures 5E - I merely demonstrate that estrogen receptor regulates the expression of two splicing factors, SRSF1 and SRSF7, in an estradiol-dependent manner. However, this finding does not support the notion that this regulation actually contributes to sex-dimorphic alternative splicing changes during human aging. Notably, the authors do not provide evidence that SRSF1 and SRSF7 expression changes actually occur in a sex-dependent manner with human aging (in a manner similar to TIA1). As such, this experimental dataset is disconnected from the main focus of the study and does not substantiate the conclusions on sex-dimorphic splicing during human aging. The authors performed RNA-seq in wild-type and ER mutant cells, and they should perform a comprehensive analysis of ER-dependent alternative splicing and compare the results with the GTEx data. It should be straightforward.

References:

Ferreira PG, Muñoz-Aguirre M, Reverter F, Sá Godinho CP, Sousa A, Amadoz A, Sodaei R, Hidalgo MR, Pervouchine D, Carbonell-Caballero J et al (2018) The effects of death and post-mortem cold ischemia on human tissue transcriptomes. Nature Communications 9: 490.

Shen X, Wang C, Zhou X, Zhou W, Hornburg D, Wu S, Snyder MP (2024) Nonlinear dynamics of multi-omics profiles during human aging. Nature Aging.

Wucher V, Sodaei R, Amador R, Irimia M, Guigó R (2023) Day-night and seasonal variation of human gene expression across tissues. PLOS Biology 21: e3001986.