Introduction

All individuals carry a minimum of 1 copy of chromosome X. Despite being a relatively long chromosome with more than 150 million base pairs [1], similar in length to chromosome 8, association testing of genetic variants on chromosome X is still not routinely conducted. Genome-wide association studies (GWAS) have been used to identify a vast range of genomic loci of interest for a variety of complex human diseases and traits by quantifying genetic variants that are statistically associated with a given disease/trait [2,3]. However, a lack of testing for variants on the X chromosome limits our ability to identify vital loci and subsequently understand potential mechanisms linked to this chromosome. There was a call for the inclusion of chromosome X into genome-wide association analyses presented in 2013. At that time, a scan of published GWAS from 2010 and 2011 showed that only 33% of the studies had tested variants on the X chromosome in their analyses [4]. Despite this call for inclusion, the lack of representation of this chromosome has not improved according to a 2023 study. Of the 136 publications that submitted at least 1 summary statistics file to the NHGRI-EBI GWAS Catalog in 2021, only 25% reported chromosome X results [5].

Indeed, there are several characteristics of this chromosome that make it unique compared to the autosomes, which can pose analytical challenges in association testing. Such challenges include how to account for X inactivation in individuals with an XX karyotype, how to model the hemizygous state of genotypes in individuals with an XY karyotype, or how to best code genotypes at the 2 pseudo-autosomal regions, short stretches at either end of the X with high homology with the Y chromosome, known as PAR1 and PAR2. The non-pseudo-autosomal region (nonPAR) denotes the middle sequence of the X chromosome.

Furthermore, there are many well-used software that take GWAS summary statistics as input and ignore chromosome X information [6,7]. This practice can make it difficult and unintuitive for researchers to run association testing on the X chromosome.

Inclusion of chromosome X routinely in GWAS and downstream analyses will serve to enhance our understanding of the genetic contributors to complex diseases and traits. Here, we propose 8 tips to help move towards the inclusion of X in GWAS to provide a suggested set of concrete actions that can be taken to overcome the challenges or obstacles preventing routine analysis of this chromosome.

Tip 1: Retain chromosome X variants in your quality control pipeline

Chromosome X variants tend to have lower quality whether genotyped, sequenced, or imputed compared to autosomal variants [8,9]. Perform careful quality control checks genome-wide, including variants on the X chromosome instead of discarding genotyping or sequencing calls for variants on the X chromosome. For sequencing data, depth on the X can be used to get a sense of each study participant’s sex chromosome karyotype. For genotyping array data, F statistics of X chromosome heterozygosity, as implemented in PLINK, for example, can be estimated to accomplish this task. Chromosome X-specific variant filters as reviewed by Keur and colleagues [10] should be incorporated, including testing for missingness by sex, call rate by sex, and the proportion of heterozygotes in individuals with XY karyotypes should be assessed. Additionally, Hardy–Weinberg Equilibrium models accounting for different ploidy on chromosome X (for example, diploid for females and haploid for males in the nonPAR region) with different assumptions with regard to X inactivation (see Tip 6) have been proposed [11].

Tip 2: Impute chromosome X variants

For genotyping array data, maximize the number of genetic variants available for association testing by imputing genotypes of variants that are not present on the array, including variants on the X chromosome. Various free online resources can be used to carry out the computationally intensive task of genotype imputation using available whole-genome sequencing imputation panels (Table 1), including the Trans-Omics for Precision Medicine (TOPMed) Imputation Server [12], Sanger Imputation Service [13], and Michigan Imputation Server [14]. The TOPMed and Michigan Imputation Servers will perform some basic chromosome X-specific quality control on the uploaded genotypes, including verifying that all variants in the nonPAR region are either haploid or diploid. For these 2 servers, chromosome X is split into 3 parts (PAR1, nonPAR, and PAR2) for statistical phasing and genotype imputation, and then subsequently merged. In contrast, the Sanger Imputation Service does not perform quality control steps on the genotypes. A thorough assessment of the various imputation panels and tools available to handle chromosome X imputation is pertinent but is beyond the scope of this short educational article.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive summary of some available genotype imputation panels.

https://doi.org/10.1371/journal.pcbi.1012160.t001

Tip 3: Use genetic association software that supports X chromosome testing

It is necessary to understand how conventional genetic association study software handles the X chromosome because different tools handle association testing on the X chromosome differently (Table 2). These differences can cause some missteps for researchers trying to include the X chromosome in their analyses for the first time. While there is no one recommended tool or approach for X chromosome association analysis, one study observed that in samples with a skewed male:female ratio, coding the male chromosome X genotypes as 0/2 (rather than 0/1) alleviated type 1 error [15]. Therefore, when male:female imbalance is present in an analysis dataset, coding of male genotypes on the X should be considered. We recommend implementing accepted genome-wide significant thresholds that account for the multiple testing inherent to genome-wide scans (for example, 5 × 10⁻⁸, which adjusts for 1 million independent tests) [16], rather than deriving X-specific significance thresholds.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Non-comprehensive list of software for association testing, including chromosome X variants.

Flags for single variant association testing are described.

https://doi.org/10.1371/journal.pcbi.1012160.t002

Tip 4: Perform stratified association analyses for X chromosome variants

Stratified analyses carrying out association tests in individuals with XX karyotypes separately from those with XY karyotypes can be conducted to identify associations on the X (as well as for autosomal variants) that differ in magnitude, direction, or significance depending on sex chromosome karyotype. Tests of heterogeneity to identify genetic variants that exhibit significant association differences in the stratified association tests should also be carried out to quantity possible association differences. As sample sizes increase, individuals with sex-chromosome aneuploidy should be included in analyses, possibly stratified according to sex chromosome karyotype. However, stratifying study individuals into subgroups will reduce the total sample size for the association test. In datasets where statistical power is a concern, there are alternative methods to test for effect of sex, such as testing for genetic variant-by-sex interactions.

Tip 5: At the minimum, assess biallelic single-nucleotide polymorphisms and variants as well as small insertions and deletions (indels) on chromosome X

Limited work has been done on investigating genetic variant–trait associations for more complex types of genetic variation on the X chromosome. Nevertheless, there is evidence of many types of genetic variation on the X chromosome contributing to complex disease risk, thus warranting additional investigations across traits and across types of variation (Table 3). When possible, expand to other variation types such as copy number and structural variants and include variants with more than 2 alleles (i.e., multi-allelic variants).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of selected types of genetic variation with known associations for complex human diseases or traits on chromosome X.

https://doi.org/10.1371/journal.pcbi.1012160.t003

Tip 6: Consider implementing various statistical models to account for X inactivation

In XX individuals, there can be compensatory mechanisms to reduce the dosage of gene expression in which one copy is subject to inactivation, resulting in only one copy of the gene being expressed (Fig 1). This biological factor must be considered while performing genetic association analyses on chromosome X. Statistical methods have been proposed to account for different inactivation models: random, skewed, or escape from inactivation [23,24,31,32]. PLINK [17,18] and XWAS [23] can model escape from inactivation or random inactivation, and there are models for skewed inactivation [31,33]. Given the lack of widely used standards and guidelines for association testing on the X chromosome, it can be useful to report association results from more than one applied method to further research in this direction using the most recent map of X inactivation across human tissues [34]. Applying and comparing association models accounting for different models of X inactivation for genetic variants on the X chromosome in XX individuals will be an important step forward to foster chromosome X association testing.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Representation of proposed X inactivation models in XX individuals.

In random X inactivation, 50% of cells have 1 allele inactive, whereas the other 50% have the other allele inactive (a). In skewed (also known as nonrandom) inactivation, more than 50% of cells (say more than 75% or more extreme) have the same allele inactive (b). Finally, escape of X inactivation describes the scenario where both alleles are active in all cells (c).

https://doi.org/10.1371/journal.pcbi.1012160.g001

Tip 7: Make your code and association summary statistics, including chromosome X results, publicly available

Data availability increases transparency and reproducibility and facilitates future research within the broader research community. Share your results with the broader community in a peer-reviewed publication describing your work. Sharing of summary statistics is commonplace and now required by many journals for publication of primary research articles. These summary statistics are used by the broader research community for a variety of tasks including replication of association signals, meta-analysis of cohorts in which the same or similar trait is measured, carrying out two-sample mendelian randomization, identification of variants to include in the development of a polygenic score and their corresponding weights, and many more downstream analyses. Not to mention, including chromosome X in summary statistics facilitates the inclusion of variants on this chromosome in these downstream analyses, furthering knowledge on the contribution of chromosome X variants to complex traits and diseases.

Code availability goes hand in hand with data availability in the call for open science. Making code available can provide other researchers with a starting point to facilitate the inclusion of X variants in their analyses. To this end, it is important to provide annotated code where the implemented steps are well documented. Guidelines described elsewhere should be followed to ensure code reproducibility that supports open science [35,36].

Tip 8: Embrace the biological complexity and value of X chromosome analyses

By following these tips, we can all do our part in increasing the inclusion of chromosome X in association analyses and embracing our biological complexity through association analyses.

Conclusions

Testing variants on the X chromosome is important for increasing understanding of this chromosome’s role in complex traits and diseases. Not to mention, making these results available to the broader community will facilitate downstream analyses including meta-analyses, the creation of polygenic risk scores and causal inference using mendelian randomization that incorporate chromosome X to advance the field. That being said, the inclusion of X chromosome genetic variants in association analyses is not a “one size fits all” approach. There is no single software or X inactivation model that will always give optimal results for every trait and in every study design. We encourage researchers to conduct further investigations to choose the best tool, resource, or method for their specific task using the information provided here as a starting point. All in all, addressing the challenges associated with chromosome X analyses will be crucial to foster future opportunities for scientific discovery related to topics that are not yet well understood, including, but not limited to, understanding haploid versus diploid states, sex chromosome aneuploidy states, and the biology of nonrandom X inactivation.