Using human-specific genomic changes to inform studying human-specific features has been a hot topic (Pollen et al., 2023) since the first discovery of human accelerated regions (HARs) that were highly conserved across vertebrates but with an accelerated rate of evolution from chimpanzees to humans in nucleotide substitution (Pollard et al., 2006). The roles of HARs have been confirmed in regulation of human-specific morphological features such as limb and brain development (Prabhakar et al., 2008; Boyd et al., 2015). Additional human-specific genomic changes have been uncovered including human accelerated DNase I hypersensitive sites (haDHS) (Gittelman et al., 2015), human biased cis-regulatory elements (Prescott et al., 2015), Hominin-specific gene regulatory elements (hsGRE) (Castelijns et al., 2020), and large structural changes such as human-specific insertions (hsInsert) that tend to be fixed in human populations (McLean et al., 2011; Kronenberg et al., 2018).

Furthermore, researchers have uncovered substantial genomic changes that are evolutionarily specific to primates (Tay et al., 2009; Vermunt et al., 2016). These primate-specific genomic information (PSI), including the human-specific genomic changes, form an important resource that could potentially shed light on complex mechanisms regulating human phenotypes (Juan et al., 2023). For example, inactivation of the uricase gene in higher primates has led to a much higher base level of urate concentration than that in species with intact uricase such as mice, and thus may have shaped a special urate-mediated physiology (Lu et al., 2019). Indeed, PSI elements were found in regulatory regions of urate associated genes (Takei et al., 2021) and in the human uricase gene UOX (Castelijns et al., 2020).

There are however a few issues limiting PSI applications. First, heterogeneity among PSI elements identified in different studies using evolved definitions and small samples. For example, overlaps exist in only a few of ∼2,700 HARs reported from various studies (Levchenko et al., 2018). Second, there is dearth of functional information for the vast majority of the PSI elements, which often requires right tissues or cell types to dissect (Doan et al., 2016; Parikshak et al., 2016; Won et al., 2019) and thus is rather expensive and time consuming. Third, a clear profile of the PSI elements and their interrelations is lacking, e.g., genomic distribution, frequencies in populations.

Another difficulty is to clarify potential roles of PSI elements in regulating human diseases, given that PSI elements are concerned at the species level rather than within-species or individual levels. PSI elements tend to be subject to positive selection leading to fixation in human populations (McLean et al., 2011; Kronenberg et al., 2018) while such positive selection could be reverted to negative selection during human evolution (e.g., adaptation to different environments) (Doan et al., 2016; Levchenko et al., 2018), leading to phenotypic variations including diseases. Furthermore, mutations can happen in PSI elements where deleterious variants may be associated with diseases such as neural disorders underlying the emergence of the human brain (Doan et al., 2016; Castelijns et al., 2020; Takei et al., 2021). These attributes together render PSI elements as favorable targets for new insights into disease associations and primate-specific regulatory mechanisms.

Here we characterize PSI elements collated from a range of publications mostly on brain-related features and their colocalization with common variants identified in genome-wide association studies (GWASs). We quote genomic positions in GRCh37/hg19 throughout.

To prepare for a whole genome sequencing based project aimed to define human specific genetic variants in brain developmental disorders, journal articles studying primate-specific and/or human-specific genomic features were reviewed and selected based on relevance and usefulness for annotating sequencing variants. We then extracted data from each selected journal article and converted the genome coordinates to GRCh37/hg19 when necessary.

The PSI elements collated were first sorted by chromosome locations and merged into PSI regions without overlaps with each other (Supplementary Table S1). Subsequently, distribution and summary statistics of PSI regions were generated. The PSI regions were then used as a map to examine colocalization of common variants identified in GWASs of 10 quantitative traits and 15 dichotomous disease traits outlined in Supplementary Table S2. Mapping GWAS variants to PSI regions and statistical tests were performed using R (R Core Team. R, 2018) (https://www.R-project.org, v4.1.1). Chi-square tests were used to examine hypothesis that no difference between the expected and the observed frequencies in one or more categories of a contingency table, or hypothesis of independence between two distributions. For each GWAS, genome-wide significant (p < 5 × 10⁻⁸ or log₁₀Bayes Factor >6 as appropriate in the study) variants were extracted and each tested for mapping to the PSI regions and recorded with the PSI type if mapped using a self-developed R script (available on request). Then the numbers of independent loci were derived from the input and the mapped variants respectively by counting linkage disequilibrium blocks (r ² > 0.8) associated with the lead associated variants that were mutually independent and reported in the original paper. For example, if five mapped variants are within a linkage disequilibrium block of a GWAS lead associated variant, they will be counted as one mapped locus. Mapping rate was then calculated as percentage of the number of mapped loci out of the number of GWAS loci. The GTEx Portal (Lonsdale et al., 2013) (https://gtexportal.org/home/) was used to retrieve functional information such as expression quantitative trait loci (eQTL) or splicing quantitative trait loci (sQTL) for variants of interest.

We studied 19,767 PSI elements identified from nine studies each with a different PSI type based on DNA sequences and/or brain related tissues or cell types (Table 1). These PSI elements formed 19,473 non-overlapping PSI regions that are generally small (Supplementary Table S1), with an average length of 1,268 base pairs, together cover only 0.81% of the genome. Only 226 PSI regions (1.2%) were derived from multiple PSI elements confirming fairly limited overlaps between PSI types. The PSI regions distributed unevenly by chromosome (p < 2.2 × 10⁻¹⁶, Chi-squared test accounting for chromosomal lengths), where the chromosomal length coverage by PSI regions fluctuated around 0.8% but was above 1% in chromosomes 11 and 20 and as low as 0.5% in chromosomes 21, 22, and X (Figure 1A).

Of the 19,473 PSI regions, 493 (∼2.5%) were mapped by 1,499 (representing 468 independent loci) out of 253,200 genome-wide significant variants (representing 5,701 independent loci) identified in GWASs of 10 quantitative and 15 dichotomous disease traits, giving a mapping rate of 0.6% (or 8.2%) at the variant (or locus) level (Supplementary Table S2). Of the 1,499 mapped variants, 332 (∼22%) were pleiotropic and associated with multiple traits (Supplementary Table S3). At the variant level, the top mapped PSI types were hsGRE, psCRE, hsInsert, and HAR (Figure 1B), where hsInsert was the only one mapped with more disease variants than quantitative trait variants and was significantly different from the remaining (p = 9.3 × 10⁻¹², Chi-squared test). At the locus level, disease loci were more likely mapped to PSI regions than quantitative trait loci (p = 1.6 × 10⁻⁰⁵, Chi-squared test) (Figure 1C). Considering hsInsert and psTU as the only PSI types without obvious regulatory evidence, a further locus level comparison suggested that the mapped GWAS loci were highly biased towards regulatory regions (p = 1.4 × 10⁻⁶⁶, Chi-squared test) (Figure 1D). Nevertheless, non-regulatory PSI regions colocalized with 23.9% (118 out of the 493) of the mapped GWAS loci and thus played an important part as well.

Figure 2 showed an exemplar region within the LRP4 gene harboring GWAS variants and rare pathogenic variants as well as a psTU overlapped with a hsInsert downstream and another hsInsert upstream. The region also harbors brain related functional elements including human gain enhancers (Reilly et al., 2015), differentially expressed long non-coding RNA (Parikshak et al., 2016), interacting transcription start sites (Won et al., 2016), and three ancient enhancers in neocortex shared across mammals (Emera et al., 2016).

We presented a profile of PSI elements distributed across the genome for the first time. These PSI elements jointly covered only 0.81% of the genome space but colocalized with GWAS common variants in 2.5% of the non-overlapping PSI regions derived. Disease variants appeared to be more likely to locate in a PSI region than quantitative trait variants, particularly in those without obvious regulatory roles (Figure 1). We further showed in the LRP4 exemplar region the colocalization of PSI elements with common and rare variants as well as additional functional elements (Figure 2) signaling complex regulatory mechanisms. These results together render PSI elements as a valuable source to inform genetic studies of complex diseases.

It is not unexpected that majority of the mapped PSI regions are regulatory because GWAS variants tend to appear in regulatory regions (Edwards et al., 2013; Buniello et al., 2019). However, since PSI regulatory elements conveying evolution perspectives, they can bring new insights into complex regulation mechanisms as shown in the LRP4 example here and previous examples of SLC2A9 regulating urate levels (Takei et al., 2021) and CACNA1C regulating Bipolar Disorder and Schizophrenia (Song et al., 2018). On the other hand, non-regulatory PSI regions are clearly involved in various phenotypes particularly diseases. Considering the downstream hsInsert in the LRP4 example (Figure 2), within it we did find rs186930464, instead of being fixed in human genome (Kronenberg et al., 2018), with varied allele frequencies in the 1,000 Genomes project and strong eQTLs and sQTLs in multiple tissues (https://gtexportal.org/home/snp/rs186930464) (Supplementary Tables S4, S5). Integrating existing functional information of its overlapped regulatory elements could better our understanding of the hsInsert and other PSI elements that were often derived from small samples with limited replication. Further functional investigation is needed to understand how these “non-regulatory” PSI elements work in shaping disease phenotypes.

The observed colocalization of PSI regions with GWAS common variants can be regarded as new evidence supporting the hypothesis of speciation characterized by substantial rewiring of the regulatory circuitry of functional genes (Castelijns et al., 2020), suggesting that every human trait would have some human-specific characteristics, as previously showed in immune and metabolic conditions (O’Bleness et al., 2012). Since PSI elements are shared across human populations, they are likely associated with common diseases but unlikely with rare diseases.

This pilot study is limited by incomprehensive inclusion of PSI elements and incomprehensive analysis scenarios. Caution is therefore recommended when interpreting the results. Further investigations are warranted to comprehensively explore PSI values in informing genetic studies of complex diseases.