MicroRNA-eQTLs in the developing human neocortex link miR-4707-3p expression to brain size
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
This paper is one of the first demonstrations that expression quantitative trait loci (eQTL) affecting human microRNAs are linked to brain development affecting brain structure and function. These findings will have a broad impact on the genomics, neural development, and microRNA fields. The datasets produced here (developmental changes in miRNAs, new human miRNAs) will likely be used for further discoveries. However, some claims need to be tempered.
(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 #1 agreed to share their name with the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Genetics and Genomics (eLife)
Abstract
Expression quantitative trait loci (eQTL) data have proven important for linking non-coding loci to protein-coding genes. But eQTL studies rarely measure microRNAs (miRNAs), small non-coding RNAs known to play a role in human brain development and neurogenesis. Here, we performed small-RNA sequencing across 212 mid-gestation human neocortical tissue samples, measured 907 expressed miRNAs, discovering 111 of which were novel, and identified 85 local-miRNA-eQTLs. Colocalization of miRNA-eQTLs with GWAS summary statistics yielded one robust colocalization of miR-4707–3p expression with educational attainment and brain size phenotypes, where the miRNA expression increasing allele was associated with decreased brain size. Exogenous expression of miR-4707–3p in primary human neural progenitor cells decreased expression of predicted targets and increased cell proliferation, indicating miR-4707–3p modulates progenitor gene regulation and cell fate decisions. Integrating miRNA-eQTLs with existing GWAS yielded evidence of a miRNA that may influence human brain size and function via modulation of neocortical brain development.
Article activity feed
-
-
-
Author Response
Reviewer #2 (Public Review):
The majority of genetic effects discovered in genome-wide association studies (GWAS) of common human diseases point to non-coding variants with putative gene regulatory effects. In principle, studying genetic effects on gene expression phenotypes, as mediators between genotype and disease, can help understand the underlying function of GWAS variants.
Lafferty et al., set to study the regulation of microRNA (miRNA) levels in mid-gestation human neocortical tissues as a potential contributor to brain-related phenotypes. To this end they performed miRNA expression profiling via small-RNA sequencing, followed by assaying expression quantitative trait loci (eQTLs) that locally regulate miRNA genes.
In addition to reporting some properties of miRNA-eQTLs, e.g., their tissue-specificity, the …
Author Response
Reviewer #2 (Public Review):
The majority of genetic effects discovered in genome-wide association studies (GWAS) of common human diseases point to non-coding variants with putative gene regulatory effects. In principle, studying genetic effects on gene expression phenotypes, as mediators between genotype and disease, can help understand the underlying function of GWAS variants.
Lafferty et al., set to study the regulation of microRNA (miRNA) levels in mid-gestation human neocortical tissues as a potential contributor to brain-related phenotypes. To this end they performed miRNA expression profiling via small-RNA sequencing, followed by assaying expression quantitative trait loci (eQTLs) that locally regulate miRNA genes.
In addition to reporting some properties of miRNA-eQTLs, e.g., their tissue-specificity, the authors searched for potential overlap or "colocalization" between these eQTL loci and GWAS loci for several putatively brain-related phenotypes. They reported colocalization at the locus containing the SNP rs4981455 which is an eQTL for miR-4707-3p and is also associated with global cortical surface area (GSA) and educational attainment phenotypes in GWAS. They further showed that exogenously increased expression of miR-4707-3p in primary human neural progenitor cells (as a model to study neurogenesis) derives an increased rate of proliferation.
The reported results are interesting and important, particularly for the understanding of miRNA biology. That said, as I detail below, the claim that miR-4707-3p expression modulates brain size and thus cognitive ability, although potentially consistent with the data, is not unequivocally supported by the analyses. As such, considering the potential social impact of the misinterpretations of these results, I believe the authors should explicitly discuss caveats, alternative explanations consistent with the data, and broader implications:
We thank the reviewer for their positive evaluation of our work and detailed comments. We agree that misinterpretation of our results could have negative social impacts, and now have added caveats and alternative explanations to our discussion section.
- The colocalization analysis used effectively tests whether miRNA-eQTL and GWAS variants are in linkage disequilibrium (LD), and does not formally test whether the miRNA-eQTL and GWAS signals are explained by the same genetic variant which is necessary for establishing causality. In this study, a formal test of colocalization is challenging given that the LD patterns in the eQTL data (from mixed ancestries) are dissimilar to the GWAS data (from European-descent samples). Furthermore, even if GWAS and miRNA-eQTL signals are explained by the same variant, this could be due to confounding (a confounder affecting both), or pleiotropy (genotype independently affecting both), and not necessarily that the miRNA-eQTL signal mediates the GWAS signal. These are also true for colocalization analyses of miRNA-eQTLs with mRNA-eQTLs or splicing-QTLs. One practical suggestion is whether authors can perform the colocalization analysis better, e.g., with methods such as SMR (https://yanglab.westlake.edu.cn/software/smr/#Overview).
As the reviewer mentioned, testing colocalized genetic signals using the eQTL dataset presented in this study remains challenging given the mixed-ancestry of the samples. We believe our primary test for colocalization, conditioning the miRNA-eQTL association using a secondary signal index variant, is sufficient evidence for a shared genetic signal (Nica et al., 2010). This is particularly true when looking for colocalizations between the miRNA-eQTLs and mRNA-e/sQTLs because both datasets used largely the same samples for expression quantification. However, the colocalization between the miRNA-eQTL for miR-4707-3p expression and the GWAS signal for educational attainment warrants greater scrutiny because the GWAS signal was discovered in European-descent samples.
To address this concern, we have conducted an additional colocalization test using the SMR and HEIDI methods as suggested by the reviewer (Zhu et al., 2016). We have updated the results section, “Colocalization of miR-4707-3p miRNA-eQTL with brain size and cognitive ability GWAS”:
"In addition to the HAUS4 mRNA-eQTL colocalization, the miRNA-eQTL for miR-4707-3p expression is also co-localized with a locus associated with educational attainment (Figure 5A)(2). Conditioning the miR-4707-3p associations with the educational attainment index SNP at this locus (rs1043209) shows a decrease in association significance, which is a hallmark of colocalized genetic signals (Figure 5-figure supplement 2A)(58,59). Additionally, the significance of the variants at this locus associated with miR-4707-3p expression are correlated to the significance for their association with educational attainment (Pearson correlation=0.898, p=5.1x10-7, Figure 5-figure supplement 2B). To further test this colocalization, we ran Summary-data-based Mendelian Randomization (SMR) at this locus which found a single causal variant to be associated with both miR-4707-3p expression and educational attainment (p=7.26x10-7)(60). Finally, the heterogeneity in dependent instruments test (HEIDI), as implemented in the SMR package to test for two causal variants linked by LD, failed to reject the null hypothesis that there is a single causal variant affecting both gene expression and educational attainment when using the mixed-ancestry samples in this study as the reference population (p=0.159). The HEIDI test yielded similar results when estimating LD with 1000 Genomes European samples (p=0.120). All this evidence points to a robust colocalization between variants associated with both miR-4707-3p expression and educational attainment despite the different populations from which each study discovered the genetic associations."
To strengthen the argument for colocalization, we added Figure 5-figure supplement 2.
Given the unique problem of colocalizing genetic signals from datasets with different LD patterns, we also attempted to colocalize the miRNA-eQTL for miR-4707-3p and educational attainment GWAS using eCAVIAR and coloc (Hormozdiari et al., 2016; Wang et al., 2020). Neither of these methods produced a significant colocalization between these two genetic signals at this locus. However, neither of these methods were designed or tested using mix-ancestry reference populations, and therefore we are still confident in declaring a shared genetic signal at this locus.
- Although possible, there is no direct evidence that the GWAS signals at rs4981455 for educational attainment and GSA are driven by variation in miRNA levels in the studied tissue. As the authors noted, rs4981455 is also an eQTL for the gene HAUS4. Furthermore, rs4981455 is a significant e/sQTL across almost all adult tissues in GTEx, and so likely has regulatory activity across wide ranges of cell or tissue types. Therefore, pinpointing the causal contexts mediating the effect in GWAS is impossible with the current data.
We agree that fully understanding the causal relationship, or mechanism, between rs4981455 and educational attainment is impossible with the current data. However, we believe the miRNA-eQTL at rs4981455, discovered in developing brain tissue, provides clues as to the causal context of this locus on educational attainment. We have updated the language throughout the manuscript to temper the notion that expression differences in miR-4707-3p is causal for changes in educational attainment (discussed below), yet we maintain that the evidence provided is consistent with miR-4707-3p playing a role in brain development ultimately leading to changes in adult educational attainment. The updated hypothesized causal relationship is shown in Figure 6H and expanded discussion on the caveats of this study, addressed in the next section, also serve to mitigate this concern.
- Orthogonal to the issues above, the genotype-to-phenotype pathway as hypothesized, i.e., genotype → miRNA levels → brain structure → educational attainment, is oversimplistic and rests on an implicit prior belief that genetic associations with educational attainment can be trivially mapped to fundamental brain features that determine cognitive ability. To illustrate the problem with this prior I refer to an old example by Christopher Jencks: in a society that prevents red-hair kids to go to school, genetic effects on hair color would be associated with educational attainment, despite having no intrinsic biological relationship with cognition. I give two scenarios consistent with the specific case of rs4981455 that are fundamentally different from what is implied in the paper: (i) The case of indirect genetic effects (see Kong et al., Science 2018). In this case, rs4981455 affects the nurturing behavior of an individual's parents, which in turn influences the individual's educational achievements and consequently brain structure features. (ii) The case of confounding. In this case, the genetic effects on brain structure are shared with another feature, such as facial shape (see Naqvi et al., Nature Genetics 2021). Variation in facial shape in a discriminatory educational environment can covary with educational attainment.
The causal pathway presented in the original version of this manuscript was indeed too simplistic and inferred a causal pathway between rs4981455 and educational attainment that was not fully backed by our data nor could be fully proved experimentally. The point we had hoped to make, and which is better represented by the updated version of Figure 6H, is that if there is a causal relationship between rs4981455 and educational attainment mediated by miR-4707-3p expression, we may be able to detect the influence of miR-4707-3p on a cellular phenotype that would explain the association of rs4981455 with cortical surface area, intracranial volume, and head size.
An updated discussion summarizes how we were not able to find evidence for a molecular mechanism consistent with the radial unit hypothesis, but that a biological link between the miRNA-eQTL and GWAS phenotypes may yet be uncovered:
"We did find one colocalization between a miRNA-eQTL for miR-4707-3p expression and GWAS signals for brain size phenotypes and educational attainment. This revealed a possible molecular mechanism by which genetic variation causing expression differences in this miRNA during fetal cortical development may influence adult brain size and cognition (Figure 6H). Experimental overexpression of miR-4707-3p in proliferating phNPCs showed an increase in both proliferative and neurogenic gene markers with an overall increase in proliferation rate. At two weeks in differentiating phNPCs, we observed an overall increase in the number of cells upon miR-4707-3p overexpression, but we did not detect a difference in the number of neurons at this time point. Based on the radial unit hypothesis (26,73), we expected to find an overall decrease in proliferation or increase in neurogenesis upon miR-4707-3p overexpression which would explain decreased cortical surface area. However, our in vitro observations with phNPCs do not point to a mechanism consistent with the radial unit hypothesis by which increased miR-4707-3p expression during cortical development leads to decreased brain size. This has also been seen in similar studies using stem cells to model brain size differences linked with genetic variation (74). Nevertheless, the transcriptomic differences associated with overexpression of miR-4707-3p in differentiating phNPCs suggest this miRNA may influence synaptogenesis or neuronal maturation, but these phenotypes may be better interrogated at later differentiation time points, by jointly expressing HAUS4 and mir-4707, or with assays to directly measure neuronal migration, maturation, or synaptic activity."
We believe the two cases addressed by the reviewer of indirect genetic effects and confounding which may actually explain the association between rs4981455 and educational attainment are less likely to be influencing the miRNA expression of miR-4707-3p measured in developing cortical tissue. This is combined with a discussion on the caveats of our findings and is addressed in the next section.
- The paper lacks a discussion on caveats to protect against potential misinterpretation of findings, especially considering the troubled history of linking facial shape and head morphology to human behavior and intelligence. I refer to the last paragraph of Naqvi et al., Nature Genetics 2021, as an example of such discussion. This is particularly crucial given that the frequency of rs4981455 varies across human populations. For example, it is important to state that the GSA and education attainment GWAS findings are in individuals of European descent, and may not necessarily point to an effect in other ancestries or even in European-descent individuals that differ from the GWAS samples in various ways, e.g., socioeconomic status (see Mostafavi et al., eLife 2020). In other words, these findings pertain to variation within the studied samples. On this note, it is important to state the amount of variation in multiple phenotypes explained by rs4981455 (which is likely tiny), and that it by no means determines the phenotype.
We have added a paragraph to the discussion highlighting the caveats of our analysis and protecting from overinterpretation of our findings:
"Here we have proposed a biological mechanism linking genetic variation to inter-individual differences in educational attainment. Given the important societal implications education plays on health, mortality, and social stratification, a proposed causal mechanism between genes and education warrants greater scrutiny (75,76). Any given locus associated with educational attainment may be driven by a direct effect on brain development, structure, and function, an indirect genetic effect such as parental nurturing behavior, or confounding caused by discriminatory practices or societal biases (77,78). Given that expression was measured in prenatal cortical tissue, where confounding societal biases are less likely to drive genetic associations and that experimental overexpression of miR-4707 affected molecular and cellular processes in human neural progenitors, the evidence at this locus is consistent with a direct effect of genetic variation on brain development, structure, and function rather than being driven by confounding or indirect effects. However, there are some important caveats to this statement. First, our study only provides evidence for the direct effect on the brain at this one educational attainment locus. Our study does not provide evidence for the direct brain effects of any other locus identified in the educational attainment GWAS. Second, common variation at this locus explains a mere 0.00802% of the variation in educational attainment in a population, so this locus is clearly not predictive or the sole determinant of this phenotype. Third, the GWAS for educational attainment and brain structure were conducted in populations of European ancestry, and allele frequency differences at these loci cannot be used to predict differences in educational attainment or brain size across populations. Finally, though both experimental and association evidence suggests a causal link between this locus and educational attainment mediated through brain development, we are unable to directly test the influence of miR-4707-3p expression during fetal cortical development on adult brain structure and function phenotypes. Therefore, we cannot rule out the possibility that the causal mechanism between rs4981455 and adult cognition may be a result of genetic pleiotropy rather than mediation at this locus. Despite these caveats, identifying the mechanisms leading from genetic variation to inter-individual differences in educational attainment will likely be useful for understanding the basis of psychiatric disorders because educational attainment is genetically correlated with many psychiatric disorders and brain-related traits (2,79)."
We hope that this paragraph contextualizes our results sufficiently to emphasize the high bar that must be surpassed to propose a biological link between a miRNA-eQTL and a risk loci for brain related traits while maintaining that we can not completely rule out the possibility of genetic pleiotropy.
- The main colocalization signal is tentatively shown for GSA. However, the authors casually refer to links with "brain size" or "head size" throughout the paper.
In addition to the locus showing a sub-genome wide significant association to global cortical surface area (GSA) presented in Figure 5, a GWAS for head size (Knol et al., 2020) and a GWAS for intracranial volume (Nawaz et al., 2022) (recently published since submitting the original manuscript) both show genomic associations at this locus for miR-4707-3p expression. The index variants for both traits colocalize with the miRNA-eQTL for miR-4707-3p and their effect directions match: alleles increasing expression of miR-4707-3p show association to decreased head size and decreased intracranial volume. For both of these studies, the summary data is not yet publicly available, preventing us from constructing plots at this locus (similar to those shown in Figure 5) or conducting additional colocalization analyses. To be more consistent throughout the paper, we have replaced many “head size” references with “brain size” when talking about this locus.
-
Evaluation Summary:
This paper is one of the first demonstrations that expression quantitative trait loci (eQTL) affecting human microRNAs are linked to brain development affecting brain structure and function. These findings will have a broad impact on the genomics, neural development, and microRNA fields. The datasets produced here (developmental changes in miRNAs, new human miRNAs) will likely be used for further discoveries. However, some claims need to be tempered.
(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 #1 agreed to share their name with the authors.)
-
Reviewer #1 (Public Review):
Understanding how changes in microRNA (miRNA) levels affect human brain development remains a crucial task that has been largely understudied. This paper is an unbiased, large-scale, important contribution to this effort. The authors first performed small RNA sequencing from mid-gestation human cortical tissue to identify expressed miRNAs. Then, they mapped 85 local-miRNA-eQTLs that rarely colocalize with mRNA-eQTLs for the host mRNA. This is a very significant discovery, with a broad impact on the field. The lack of colocalization between miRNA- and mRNA-eQTLs reinforces the theory that miRNA transcriptional mechanisms are often independent of the host gene. Interestingly, miRNA-eQTLs map less frequently to risk loci. Considering the well-known importance of miRNAs during brain development, it is likely …
Reviewer #1 (Public Review):
Understanding how changes in microRNA (miRNA) levels affect human brain development remains a crucial task that has been largely understudied. This paper is an unbiased, large-scale, important contribution to this effort. The authors first performed small RNA sequencing from mid-gestation human cortical tissue to identify expressed miRNAs. Then, they mapped 85 local-miRNA-eQTLs that rarely colocalize with mRNA-eQTLs for the host mRNA. This is a very significant discovery, with a broad impact on the field. The lack of colocalization between miRNA- and mRNA-eQTLs reinforces the theory that miRNA transcriptional mechanisms are often independent of the host gene. Interestingly, miRNA-eQTLs map less frequently to risk loci. Considering the well-known importance of miRNAs during brain development, it is likely that the expression mechanisms of miRNAs are tightly regulated, as the authors suggest. The authors then focus on one specific miRNA-eQTL that affects miR-4707-3p and show colocalization with GWAS signals for head size (smaller) and educational attainment (decreased). They hypothesize that miR-4707-3p affects brain development by altering progenitors' proliferation and they partially support the hypothesis by over-expressing miR-4704-3p in human neural progenitor cells and showing an increase in neurogenesis.
-
Reviewer #2 (Public Review):
The majority of genetic effects discovered in genome-wide association studies (GWAS) of common human diseases point to non-coding variants with putative gene regulatory effects. In principle, studying genetic effects on gene expression phenotypes, as mediators between genotype and disease, can help understand the underlying function of GWAS variants.
Lafferty et al., set to study the regulation of microRNA (miRNA) levels in mid-gestation human neocortical tissues as a potential contributor to brain-related phenotypes. To this end they performed miRNA expression profiling via small-RNA sequencing, followed by assaying expression quantitative trait loci (eQTLs) that locally regulate miRNA genes.
In addition to reporting some properties of miRNA-eQTLs, e.g., their tissue-specificity, the authors searched for …
Reviewer #2 (Public Review):
The majority of genetic effects discovered in genome-wide association studies (GWAS) of common human diseases point to non-coding variants with putative gene regulatory effects. In principle, studying genetic effects on gene expression phenotypes, as mediators between genotype and disease, can help understand the underlying function of GWAS variants.
Lafferty et al., set to study the regulation of microRNA (miRNA) levels in mid-gestation human neocortical tissues as a potential contributor to brain-related phenotypes. To this end they performed miRNA expression profiling via small-RNA sequencing, followed by assaying expression quantitative trait loci (eQTLs) that locally regulate miRNA genes.
In addition to reporting some properties of miRNA-eQTLs, e.g., their tissue-specificity, the authors searched for potential overlap or "colocalization" between these eQTL loci and GWAS loci for several putatively brain-related phenotypes. They reported colocalization at the locus containing the SNP rs4981455 which is an eQTL for miR-4707-3p and is also associated with global cortical surface area (GSA) and educational attainment phenotypes in GWAS. They further showed that exogenously increased expression of miR-4707-3p in primary human neural progenitor cells (as a model to study neurogenesis) derives an increased rate of proliferation.
The reported results are interesting and important, particularly for the understanding of miRNA biology. That said, as I detail below, the claim that miR-4707-3p expression modulates brain size and thus cognitive ability, although potentially consistent with the data, is not unequivocally supported by the analyses. As such, considering the potential social impact of the misinterpretations of these results, I believe the authors should explicitly discuss caveats, alternative explanations consistent with the data, and broader implications:
1. The colocalization analysis used effectively tests whether miRNA-eQTL and GWAS variants are in linkage disequilibrium (LD), and does not formally test whether the miRNA-eQTL and GWAS signals are explained by the same genetic variant which is necessary for establishing causality. In this study, a formal test of colocalization is challenging given that the LD patterns in the eQTL data (from mixed ancestries) are dissimilar to the GWAS data (from European-descent samples). Furthermore, even if GWAS and miRNA-eQTL signals are explained by the same variant, this could be due to confounding (a confounder affecting both), or pleiotropy (genotype independently affecting both), and not necessarily that the miRNA-eQTL signal mediates the GWAS signal. These are also true for colocalization analyses of miRNA-eQTLs with mRNA-eQTLs or splicing-QTLs. One practical suggestion is whether authors can perform the colocalization analysis better, e.g., with methods such as SMR (https://yanglab.westlake.edu.cn/software/smr/#Overview).
2. Although possible, there is no direct evidence that the GWAS signals at rs4981455 for educational attainment and GSA are driven by variation in miRNA levels in the studied tissue. As the authors noted, rs4981455 is also an eQTL for the gene HAUS4. Furthermore, rs4981455 is a significant e/sQTL across almost all adult tissues in GTEx, and so likely has regulatory activity across wide ranges of cell or tissue types. Therefore, pinpointing the causal contexts mediating the effect in GWAS is impossible with the current data.
3. Orthogonal to the issues above, the genotype-to-phenotype pathway as hypothesized, i.e., genotype → miRNA levels → brain structure → educational attainment, is oversimplistic and rests on an implicit prior belief that genetic associations with educational attainment can be trivially mapped to fundamental brain features that determine cognitive ability. To illustrate the problem with this prior I refer to an old example by Christopher Jencks: in a society that prevents red-hair kids to go to school, genetic effects on hair color would be associated with educational attainment, despite having no intrinsic biological relationship with cognition. I give two scenarios consistent with the specific case of rs4981455 that are fundamentally different from what is implied in the paper: (i) The case of indirect genetic effects (see Kong et al., Science 2018). In this case, rs4981455 affects the nurturing behavior of an individual's parents, which in turn influences the individual's educational achievements and consequently brain structure features. (ii) The case of confounding. In this case, the genetic effects on brain structure are shared with another feature, such as facial shape (see Naqvi et al., Nature Genetics 2021). Variation in facial shape in a discriminatory educational environment can covary with educational attainment.
4. The paper lacks a discussion on caveats to protect against potential misinterpretation of findings, especially considering the troubled history of linking facial shape and head morphology to human behavior and intelligence. I refer to the last paragraph of Naqvi et al., Nature Genetics 2021, as an example of such discussion. This is particularly crucial given that the frequency of rs4981455 varies across human populations. For example, it is important to state that the GSA and education attainment GWAS findings are in individuals of European descent, and may not necessarily point to an effect in other ancestries or even in European-descent individuals that differ from the GWAS samples in various ways, e.g., socioeconomic status (see Mostafavi et al., eLife 2020). In other words, these findings pertain to variation within the studied samples. On this note, it is important to state the amount of variation in multiple phenotypes explained by rs4981455 (which is likely tiny), and that it by no means determines the phenotype.
5. The main colocalization signal is tentatively shown for GSA. However, the authors casually refer to links with "brain size" or "head size" throughout the paper.
-