Broken silence: 22,841 predicted deleterious synonymous variants identified in the human exome through computational analysis

Abstract

Synonymous single nucleotide variants (sSNVs) do not alter the primary structure of a protein, thus it was previously accepted that they were neutral. Recently, several studies demonstrated their significance to a range of diseases. Still, variant prioritization strategies lack focus on sSNVs. Here, we identified 22,841 deleterious synonymous variants in 125,748 human exomes using two in silico predictors (SilVA and CADD). While 98.2% of synonymous variants are classified as neutral, 1.8% are predicted to be deleterious, yielding an average of 9.82 neutral and 0.18 deleterious sSNVs per exome. Further investigation of prediction features via Heterogeneous Ensemble Feature Selection revealed that impact on amino acid sequence and conservation carry the most weight for a deleterious prediction. Thirty nine detrimental sSNVs are not rare and are located on disease associated genes. Ten distinct putatively non-deleterious sSNVs are likely to be under positive selection in the North-Western European and East Asian populations. Taken together our analysis gives voice to the so-called silent mutations as we propose a robust framework for evaluating the deleteriousness of sSNVs in variant prioritization studies.

Keywords: Synonymous variants; deleterious; human exome; bioinformatics

Introduction

Point mutations in protein coding sequences may lead to remarkable functional changes and their severity can be classified by evaluating the extent of amino acid alterations to protein function (Cooper et al., 2010). Synonymous single nucleotide variants (sSNVs) do not alter the sequence of amino acids due to codon degeneracy, seemingly causing no change to protein function. Because of this, sSNVs are often discarded in variant prioritization pipelines (Buske et al., 2015).

The idea that sSNVs are innocuous has been recently challenged when several studies associated these variants to different diseases (Gartner et al., 2013; Bonin et al., 2016; Diederichs et al., 2016; Palagano et al., 2017). sSNVs in GWAS studies share similar likelihood and effect size to disease association as non-synonymous SNVs (Chen et al., 2010). The mechanisms by which sSNVs can cause deleterious consequences comprise a series of processes related to modulation of gene expression, such as aberrant splicing (Cartegni et al., 2002), modified mRNA stability (Nackley et al., 2006) and changes in the pace of synthesis and cotranslational folding of proteins due to codon usage bias (Hunt et al., 2014).

Studies on yeast have demonstrated that the majority of synonymous variants are strongly nonneutral and can have significant effects on gene expression levels (Shen et al., 2022), suggesting that synonymous variants may play a more important role in shaping an organism’s phenotype than previously thought. Despite the demonstrated importance of sSNVs, efforts to experimentally elucidate the functional consequences of sSNVs are scarce, especially when compared to initiatives validating non-synonymous variants (Buske et al., 2015). For such, while methods for predicting the consequence of sSNVs to protein function have been developed (Buske et al., 2015), we lack a robust framework for evaluating sSNVs deleteriousness to the benefit of variant prioritization studies.

Here, we sought out to develop a framework to assist in the deleteriousness prediction of sSNVs identified by whole exome sequencing (WES) data. We obtained candidate deleterious (detrimental) sSNVs by combining the prediction results of Silent Variant Analysis (SilVA) and Annotation Dependent Depletion (CADD). Next, we evaluated the weight of features to deleteriousness classification via Heterogeneous Ensemble Feature Selection, and comprehensively analyzed the frequency of variants and gene ontology. Finally, we evaluated if benign variants could be subject to positive selection using the Population Branch Statistics (PBS), an F_st based method.

Material and Methods

Dataset

All the data used here are publicly available on The Genome Aggregation Database (gnomAD v2.1.1) (Karczewski et al., 2020). We downloaded the variant call format (.vcf) files for all 24 human chromosomes, separately, containing data from 125,748 exomes, all mapped to the GRCh37/hg19 reference sequence. The Y chromosome was cut out from the analyses because one of the prediction softwares (SilVA) doesn’t have support for this chromosome.

Synonymous variants identification and effect prediction

Each .vcf file was used as input for SilVA (v1.1.1) (Buske et al., 2015), which identifies only synonymous variants and predicts their effects. SilVA bases its predictions on a number of features, including conservation, codon usage, splice sites, splicing enhancers and suppressors, and mRNA folding free energy. We used all variants classified as synonymous by SilVA as input to CADD (v1.4) (Rentzsch et al., 2019), which is a variant effect predictor not specific to sSNVs. CADD integrates multiple annotations into one metric by contrasting variants that survived natural selection with simulated mutations. Next, we selected only the variants classified as synonymous both by SilVA and CADD.

The next filter step involved the effect predicted for each variant. We separated our dataset into two groups: sSNVs predicted as deleterious and sSNVs predicted as benign. The CADD PHRED-like scaled score ranks a variant relative to all possible substitutions of the human genome (8.6x109) (Rentzsch et al., 2019). A PHRED-like score greater or equal to 20 indicates the 1% most deleterious, while a score greater or equal to 30 indicates the 0.1% most deleterious. For this study, variants with a PHRED-like score ≥ 15 were considered as detrimental and variants with a PHRED-like score < 10 were considered as benign, as recommended by the authors. On the other hand, SilVA classifies the variants as benign (score ≤ 0.270), potentially pathogenic (0.270 > score ≤ 0.485) and likely pathogenic (score > 0.485) based on the predicted score that ranges from 0 to 1, where close to 1 is more likely to be deleterious. Variants featuring both pathogenic SilVA classes (potentially and likely) were considered as deleterious for this work. Variants which had divergent effect prediction between SilVA and CADD were filtered out and the remaining sSNVs composed our final dataset.

Ensemble feature selection

To find out which features contributed the most to the prediction of detrimental variants, we performed a Heterogeneous Ensemble Feature Selection (EFS) on 53 features from the CADD annotations (^{Table S1}). The remaining 46 features were filtered out for either presenting more than 5% of missing values or not making sense to this specific analysis (^{Table S2}). We combined both Python3 and R programming languages to create an ensemble with four filter methods provided by FSelector R package (Romanski et al., 2021): Chi-square (Bommert et al., 2020), OneR (Holte, 1993), Gain Ratio (Quinlan, 1986), and Symmetrical Uncertainty (Bommert et al., 2020).

All the feature selection methods provide a features ranking, from the most to the least relevant feature to discriminate among classes (i.e., deleterious or benign) based on scores computed according to their particularities. Our ensemble setup combines the four different feature relevance opinions using the Borda Count method, a popular voting rule that combines preferences of multiple voters. Due to the severe imbalance between the number of deleterious and benign variants in our dataset, we performed an undersampling on the benign data, which is the majority class. We randomly sampled the benign class 100 times for the same amount of variants in the deleterious class and performed the Ensemble Feature Selection with all samples. The final ranking was a combination of all 100 rankings also using the Borda Count method.

Gene enrichment

To investigate the functions of the genes on which the detrimental sSNVs are located and the biological pathways implicated, we developed the R package called autoGO as an integrator of gene enrichment for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. The GO search depends on the package clusterProfiler (v3.18.1) (Yu et al., 2012) and the enrichment of KEGG pathways depends on the package KEGGprofile (v1.32.0) (Zhao et al, 2020). ClusterProfiler allows the user to perform the KEGG pathways enrichment as well, however, it uses a deprecated version of the KEGG database, while KEGGprofile connects with the up-to-date data available online. Both packages perform hypergeometric tests to assess the significance of the enrichment followed by false discovery rate (FDR) correction for multiple comparisons.

As the advantages over the currently available packages, autoGO was designed to deal with files containing numerous genes, working as a standalone application. AutoGO performs the analysis from a simple table containing the gene identifiers with or without expression data, generating standardized plots and tables for each input file, regardless of the source of the enriched terms. Consequently, AutoGO can optimize the efforts on the analysis of genomics data. Sources may be found at https://github.com/ldiass/MPSbase/tree/master/autoGO.

Identification of outliers

In order to evaluate the possibility of positive selection acting upon the benign variants dataset, we used the F_st based Population Branch Statistics (PBS) to detect outliers at first. Introduced by Wright (1951) as part of the F-statistics, the F_st is a descriptive measure of the differentiation prompted by important evolutionary processes such as migration, mutation and drift, between two populations (Holsinger and Weir, 2009). Since F_st is directly related to the variance in allele frequency, small F_st values indicate similarity of the allele frequencies within each population.

PBS was first described by Yi et al. (2010) in a study of heritable adaptations to extreme altitude in the Tibetan Plateau population. It is based on the premise that a gene presenting large differences in allele frequencies between two populations configures a potential target for natural selection. However, simply ranking F_st values wouldn’t tell which population was affected by selection. PBS introduces a third more distant population for a pairwise F_st comparison, yielding in the PBS value. Variants which present extreme PBS values - the outliers - configure strong candidates for positive selection in the first population. Here, we developed an R script to calculate PBS values as described by Yi et al. (2010).

Two groups of three populations were separately analyzed: East/South Asian (EAS-SAS), and North-Western/South European (NWE-SEU), with the third and more distant population being the African/African American for both groups. Population datasets were obtained from GnomAD using the database’s classification. These populations were chosen for the similarities between the natives from the same continent given by the minimal F_st values between them (EAS-SAS = 0.00108 and NWE-SEU = 0.00029), and for the blatant differences from the African/African American natives. The African/African American population is represented by 8.128 individuals, South Asian by 15.308, East Asian by 9.197, whereas South and North-Western European are represented by 56.885 individuals together.

As a support analysis, we separately ran the same groups of populations on Bayescan (v2.1) using default parameters. Just like PBS, Bayescan is an F_st based method used to identify variants under natural selection. The difference is that it uses a Bayesian model for estimation of locus-population specific F_st coefficients.

Results

The majority of human synonymous variants are predicted to be neutral

Our raw dataset consisted of 16,754,528 single nucleotide variants found in 125,748 exomes (Karczewski et al., 2020). After combining prediction results from SilVA and CADD, we obtained 1,266,032 sSNVs, corresponding to 7.56% of our raw dataset. Out of this set of sSNVs, 98.2% (1,243,191 variants) were considered benign (^{File S1}), while 1.8% (22,841 variants) were predicted to be deleterious by both predictors (^{File S2}), yielding an average of 0.18 deleterious sSNVs per exome (^{Figure S1}). In contrast to studies on yeasts (Shen et al., 2022), the majority of synonymous variants found in humans are predicted to be neutral.

CADD ranked 589,360 variants as the top 15% deleteriousness, which corresponds to its cutoff. Only 186 variants were ranked as the top 0.1% most deleterious (Figure 1A and 1B) with a PHRED-like score > 30. Out of 46,264 variants predicted as deleterious by SilVA, 10,720 sSNVs fell in the likely pathogenic category, with scores higher than 0.48, and only 77 obtained a score higher than 0.9 (Figure 1C and 1D). None of the 77 top-scoring variants from SilVA’s prediction are ranked among the 0.1% CADD variants.

Figure 1 -
Comparison between SilVA and CADD predictions. A. CADD’s prediction boxplot. In grey, the benign group, in green the top 10% most deleterious variants, in red the top 1%, and in blue the top 0,1%. The Y axis indicates the PHRED-like scaled score; B. Density plot of CADD’s prediction. The dashed red line indicates the cutoff for deleterious variants C. SilVA prediction boxplot. In gray, the likely benign group, in green, the potentially pathogenic group, in red, the likely pathogenic group. Y axis indicates the mean SilVA score; D. Pie chart of SilVA’s prediction. Same color code as Figure 1C .

The sSNV 9-139685876-G-A is the only featuring CADD’s top 5 scoring variants that was predicted as likely pathogenic by SilVA, with a 0.586 score. The remaining four variants featuring CADDs top 5 were considered potentially pathogenic by SilVA, with a score ≤ 0.350: 19-7747163-G-T, 20-35807771-G-A, 3-73047268-G-A, 9-35809402-G-C. Four out of five SilVA’s top 5 variants are included in the 1% predicted as most deleterious by CADD (Table 1).

Table 1 -
Top-scoring pathogenic variants from each predictor and their strongest gene-disease association according to MalaCards.

Predictor	Variant	CADD*	SilVA*	Gene	Disease (MalaCards ID)
CADD	3-73047268-G-A	44.00	0.299	PPP4R2	multiple cancers, SPN0461
	19-7747163-G-T	37.00	0.322	TRAPPC5	SPN4052, BNC0023, OST1104, RTN0415, multiple Cancers
	9-35809402-G-C	33.00	0.350	SPAG8 §	ACR1286, EPP0117, SHR0848
	9-139685876-G-A	32.00	0.586	TMEM141	INT0049, multiple cancers
	20-35807771-G-A	24.80	0.325	RPN2	CNG41110, ANM08011, multiple Cancers
SilVA	12-105537021-G-A	22.50	0.951	WASHC4 §	INT47412, ATS20413
	6-152631823-C-T	18.37	0.951	SYNE1 §	SPN20714, EMR01415, ART16516, SPS00817, EMR01818, JVN05019
	5-137488171-C-T	20.30	0.943	BRD8	multiple cancers
	8-96047804-G-A	22.30	0.936	NDUFAF6 §	MTC16420, FNC06621, LGH00722, PRM38423
	6-31922996-C-T	23.60	0.933	NELF-E	ATM09524

• ¹
  Spinal muscular atrophy. ²X-Linked Spondyloepiphyseal Dysplasia Tarda. ³Binocular Vision Disease. ⁴Osteogenesis Imperfecta Type Xv. ⁵Retinitis Pigmentosa 11. ⁶Acromesomelic Dysplasia 1. ⁷Epiphyseal Chondrodysplasia, Miura Type. ⁸Short Stature With Nonspecific Skeletal Abnormalities. ⁹Intraneural Perineurioma. ¹⁰Congenital Disorder of Glycosylation, Type in. ¹¹Anemia, Congenital Dyserythropoietic, Type Iiia. ¹²Autosomal Recessive Intellectual Developmental Disorder 43. ¹³Autosomal Recessive Non-Syndromic Intellectual Disability. ¹⁴Spinocerebellar Ataxia 8. ¹⁵Autosomal Dominant Emery-Dreifuss Muscular Dystrophy 4. ¹⁶Myogenic Type Arthrogryposis Multiplex Congenita 3. ¹⁷Spastic Ataxia. ¹⁸Autosomal Dominant Emery-Dreifuss Muscular Dystrophy 2. ¹⁹Juvenile Amyotrophic Lateral Sclerosis. ²⁰Mitochondrial Complex I Deficiency, Nuclear Type 17. ²¹Fanconi Renotubular Syndrome 5. ²²Leigh syndrome. ²³Primary Fanconi Renotubular Syndrome. ²⁴Autoimmune Disease.
• ^§
  Gene is likely to be associated with causing the disease(s), since their gene-disease associations are supported by manually curated and trustworthy sources.
• *SilVA classifies the variants as benign (score ≤ 0.270), potentially pathogenic (0.270 > score ≤ 0.485) and likely pathogenic (score > 0.485), whereas CADD uses a PHRED-like score system, separating the deleterious category into the 1% most deleterious (PHRED score ≥ 20) and the 0.1% most deleterious (PHRED score ≥ 30). For this study, variants with a PHRED-like score < 10 were considered as benign.

“Consequence” is the most relevant feature to the effect prediction

Using the comprehensive annotation provided by CADD for each variant (^{Tables S1} and ^S2), we performed a Heterogeneous Ensemble Feature Selection in order to rank these annotations according to their relevance to the effect prediction. The top ten most relevant features are related to the type of variant, conservation scores, chromatin state, and the reference amino acid (^{Table S1}). The most relevant feature is ”consequence”. Since we are assessing only synonymous mutations, our final dataset presented only two types of consequences: ”synonymous” and ”splice site” (synonymous variants occurring in splice sites). From the deleterious dataset, 75.72% of the variants occur in splice sites, whereas only 1.12% of benign variants occur in splice sites. Interestingly, 25% of deleterious variants occur outside splicing sites. The remaining features in the top eight are conservation scores given by different softwares. The ninth most relevant feature is the proportion of heterochromatin state in 127 cell types.

More than 90% of the sSNVs are super rare

Next, we categorized detrimental sSNVs according to their allele frequency (AF) into not rare, rare and super rare variants. In our data, more than 90% of the variants from both classes, benign and deleterious, present an AF ≤ 0.1% and are considered here as super rare (^{Figure S2}). The proportion of super rare detrimental variants is statistically higher than super rare benign variants, whereas the proportion of rare (0.1% ≥ AF ≤ 1%) and not rare (AF > 1%) deleterious variants is statistically lower than rare and not rare benign variants (p < 2.2x10−16).

Out of the 22,841, 21,926 were considered super rare deleterious variants, located on 9635 different genes. KEGG pathway (Figure 2A; ^{File S3}) and Gene Ontology (^{Figure S3}; ^{File S4}) enrichment analyses revealed that metabolic pathways are enriched in 438 genes (FDR = 2.12x10−14), corresponding to less than 35% of the total gene set where this pathway is represented. Nevertheless, the spinocerebellar ataxia pathway was found to be enriched in 66 genes (FDR = 2.76x10−9), corresponding to almost 50% of the gene set associated with this pathway.

Figure 2 -
KEGG pathways analysis. A. Top-10 KEGG pathways enriched in the super rare variants genes; B. KEGG pathways enriched in the rare variants genes.

The 876 rare deleterious variants are located on 827 different genes, yet only 63 were found to be implicated in four different KEGG pathways (Figure 2B ; ^{File S5}) according to enrichment analyses (^{Figure S4}; ^{File S6}): Ubiquitin mediated proteolysis (FDR = 0.00039, 15 genes), Lysosome (FDR = 0.00046, 14 genes), Endocytosis (FDR = 0.00056, 22 genes) and Protein digestion and absorption (FDR = 0.00055, 12 genes). The latter is the only pathway not found to be enriched in the super rare deleterious sSNVs genes.

For the 39 not rare deleterious variants, no KEGG pathway was found to be enriched in the 39 different genes on which they are located. Remarkably, all genes have been associated with one or more diseases, with the exception of GKAP1, on which the sSNV 9-86354657-A-C is located (Table 2).

Table 2 -
Not rare pathogenic variants and their strongest gene-disease association according to MalaCards.

Variant	MAF	ABraOM	SilVA	CADD	Gene	Disease (MCID†)
						Monogenic
22-38051628-C-A	0.02048	0.045230	0.479	17.10	PLCD1 §	NLD012, LKN007
11-77090939-C-T	0.01339	0.006568	0.333	15.09	PAK1 §	INT331
						Multifactorial
17-7468277-C-T	0.03548	0.022167	0.504	18.72	SENP3	PSR032, ATS523, multiple cancers
12-122292609-C-T	0.01160	0.007389	0.590	22.90	HPD §	HWK001, TYR011
17-80399692-T-G	0.02413	0.022167	0.288	18.50	HEXDC	MLD018, AMY004
10-63450379-G-A	0.02376	0.015599	0.445	21.70	CABCOCO1	CNR021
10-56106173-T-C	0.01359	0.021346	0.346	16.76	PCDH15 §	USH041, DFN093
6-46851296-C-T	0.01190	0.020525	0.400	17.54	ADGRF5	PLY117, RTN048, SFT011
20-44580788-G-T	0.02123	0.037767	0.344	22.10	ZNF335 §	MCR223
21-33678976-G-T	0.01311	0.003284	0.368	18.28	MRAP	GLC043, FML063
19-41086309-G-C	0.03815	0.042693	0.419	16.05	SHKBP1	AML050, NNN034, BRK001, DLF001, KRN001, SML004
6-31938120-C-T	0.01684	0.018062	0.362	22.60	DXO	CTS005, CRY008
11-86159223-C-T	0.05721	0.061576	0.368	17.66	ME3	THY062
12-132416780-C-A	0.01648	0.015599	0.432	16.76	PUS1 §	MYP021
15-64452322-G-A	0.01375	0.010673	0.305	18.60	PPIB §	OST130, BRT054, OST122, OST121, OST080
16-66764069-G-C	0.01689	0.011494	0.405	16.61	DYNC1LI2	BRD019
11-76867135-C-T	0.04401	0.033662	0.312	20.30	MYO7A §	USH036, DFN250, DFN251, USH001, RRG078, RTN008, SNS001, USH035, FND002, NNS072, CNR004, NNS044, ERM002, RRT027, RRT028
9-86354657-A-C	0.01474	0.002463	0.360	15.01	GKAP1	multiple cancers
16-67917958-G-A	0.03065	0.044335	0.316	22.00	EDC4	ULN001, THR013
4-113468564-C-T	0.05138	0.030378	0.309	22.20	ZGRF1	ISL163
13-30107118-A-C	0.04922	0.022167	0.359	15.66	SLC7A1	HYP595
19-39329205-C-T	0.02172	0.025452	0.352	20.90	HNRNPL	AZS001, MTH009
X-153176369-T-C	0.01101	-	0.339	18.67	ARHGAP4	NPH007, DBT005, XLN251
3-58376875-T-G	0.01735	-	0.316	15.25	PXK §	SYS001
13-51941943-T-C	0.01705	0.009852	0.317	18.81	INTS6	multiple cancers
13-46942949-A-C	0.1767	0.041051	0.656	16.39	RUBCNL	multiple cancers
15-81294774-G-C	0.05068	0.040230	0.339	15.71	TLNRD1	BLD134
18-43604634-C-T	0.01421	0.006568	0.339	15.60	PSTPIP2	CHR288, SPH001
1-54704829-T-C	0.01544	0.032841	0.397	16.85	SSBP3	LSS002, CCK001
5-90459600-T-G	0.02762	0.004105	0.388	15.77	ADGRV1 §	USH020, FBR069, USH001, USH035, FND002, RRG078, EPL140, USH037, GNR002
7-89937168-G-A	0.02964	0.008210	0.817	23.50	CFAP69 §	SPR127, NNS033
16-4487486-G-A	0.01564	0.004926	0.705	19.68	DNAJA3	CDS002, MYS074, ALT004
						Both
7-117199709-G-A	0.01672	0.027094	0.543	23.10	CFTR §	CYS001, PNC108, HRD234, BRN076, VSD002, MLN007, PRS050, SPR093, AQG005, IDP074, NCH001, MLN084
3-143371201-C-T	0.03362	0.050082	0.417	22.00	SLC9A9 §	ATS377, CLR023
19-11325229-C-T	0.01410	0.029557	0.845	23.00	DOCK6 §	ADM007, FML021
17-33477242-G-A	0.06422	0.046798	0.372	21.90	UNC45B §	CTR144, MYP004, MYF012, ERL036, ERL043
14-23856861-C-T	0.01165	0.018883	0.319	22.00	MYH6 §	CRD089, CRD096, ATR022, SCK022, CRD086, DLT002, HRT038, PTN001, CRD233, FML304, FML272
11-6650684-T-A	0.009917	0.009852	0.307	22.70	DCHS1 §	VNM003, MTR077, MTR080
15-91543131-T-A	0.01862	0.047619	0.681	22.90	VPS33B §	ART062, KRT080, CHL193

• ^†
  MalaCards ID. For the complete description of disease names, refer to ^{Table S3}.
• ^§
  Gene is likely to be associated with causing the disease(s), since their gene-disease associations are supported by manually curated and trustworthy sources.

FOXD4L5 carries the best candidate for positive selection in the NWE population

We then decided to evaluate if non deleterious sSNVs were subject to evolutionary constraints, given they comprise the majority of our data. Positive selection acting upon variants classified as benign by both SilVA and CADD (1,243,191 variants) was investigated using the population branch statistics (PBS) to pairwise compare the F_st values of variants from two groups of closely related populations against a distant one. In one group (EAS-SAS) we compared East and South Asians, and in the other group (NWE-SEU), we compared North-Western and Southern Europeans, both using African/African American as the outgroup population. PBS is an F_st-based method with good power to detect recent selection by measuring alleles with extreme frequency in a specific population when compared to two other populations (Yi et al., 2010). Variants showing extreme PBS values - the outliers - represent strong candidates for positive selection.

The top 0.1% higher F_st values between the NWE-SEU group encompassed 168 sSNVs (Figure 3A ), while 145 sSNVs were found in the top 0.1% F_st of the EAS-SAS group (Figure 3B). The two overall best outlier candidates are found to be positively selected on the North-Western European population: a C>T change (PBS = 0.42) on the gene FOXD4L5 and an A>G change (PBS = 0.35) on the gene CHRFAM7A (Table 3). The other three candidates to complete the top 5 for the North-Western European population are the T>C variant (PBS = 0.17) situated on the gene HERC2; a T>C variant found in the Nodal Modulator 3 (NOMO3) gene (PBS = 0.13); and the G>A sSNV (PBS = 0.08) located on the gene Leucine Rich Repeat Containing 37A (LRRC37A).

Figure 3 -
Manhattan plot showing PBS statistics. A. PBS outlier analysis for the NWE-SEU group. Highlighted in green, the outliers featuring the top 0.1% best candidates for positive selection. B. PBS outlier analysis for the EAS-SAS group. Highlighted in green, the outliers featuring the top 0.1% best candidates for positive selection.

Table 3 -
PBS outliers and their genes.

Group	Variant	PBS	Gene	Description
SEU-NWE	9-70177312-C-T	0.428844876	FOXD4L5	Forkhead Box D4 Like 5
	15-30654889-A-G	0.353300602	CHRFAM7A	Fusion of CHRNA7 and FAM7A
	15-28467246-T-C	0.170901254	HERC2	HECT And RLD Domain Cont. E3 Ubiq. Prot. Ligase 2
	16-16367702-T-C	0.136165561	NOMO3	NODAL Modulator 3
	17-44408795-G-A	0.082536940	LRRC37A	Leucine Rich Repeat Containing 37A
SAS-EAS	16-22547298-G-T	0.19979496	NPIPB5	Nuclear Pore Complex Interacting Protein B5
	10-47911591-T-G	0.15311104	WASHC2A	WASH Complex Subunit 2A
	16-1279253-T-C	0.11847263	TPSB2	Tryptase Beta 2
	16-81242151-T-C	0.11058199	PKD1L2	Polycystic Kidney Disease 1-Like 2
	2-107040985-T-C	0.10454195	RGPD3	RANBP2 Like And GRIP Domain Containing 3

A G>T change on NPIPB5 is a great candidate for positive selection in the EAS population

The third highest scoring sSNV overall was found to be positively selected on the East Asian population. The G>T change (PBS = 0.19) is located on the gene NPIPB5 on chromosome 16. The other four variants to complete the five best candidates to be positively selected in the East Asian population, in relevance order, are: the T>G variant (PBS = 0.15), found in the gene WASHC2A; the T>C variant (PBS = 0.11) located on the gene Tryptase β 2 (TPSB2); a T>C change (PBS = 0.11) on the gene Polycystic Kidney Disease 1-Like 2 (PKD1L2); the T>C variant (PBS = 0.10) situated on the gene RGPD3.

We ran the same groups of populations on Bayescan (v2.1) (Foll and Gaggiotti, 2008), which identifies variants under natural selection using a Bayesian model for estimation of locus-population-specific F_st coefficients. Bayescan was not able to identify any outliers (^{Figure S5}), maybe because of high F_st values and the small number of populations. This method is known for being conservative, which allows for a low false positive rate (Narum and Hess, 2011), although robustness decreases when the number of screened populations is low (Tigano et al., 2017).

Discussion

Synonymous variants are often considered to be neutral. However, they may impact phenotype through different mechanisms. For example, synonymous variants may have an impact on splicing, through splice consensus disruption or leading to alternative splicing. But, it still might be pathogenic, even if no effect on splice consensus sites or alternate splicing are predicted. However, such variants would probably be classified as Likely Benign with computational support. Moreover, if splicing impact is suspected or evidence hints at pathogenicity, the American College of Medical Genetics’ (ACMG) recommends it should be classified as uncertain until functional evaluation or further evidence is available (Richards et al., 2015).

This proposed framework holds potential in the reclassification of synonymous Variants of Uncertain Significance (VUS). Here, we leveraged comprehensive genomic data from multiple populations, allowing for a more extensive assessment of the potential impact of sSNVs. By employing this approach, our study contributes with valuable insights into the functional consequences of these variants, thereby offering a promising avenue for the reclassification of VUS. Through the integration of comprehensive population data and advanced genomic analysis, this method holds promise in enhancing our understanding of VUS and their potential association with diseases. Nonetheless, the utilization of this extensive analysis within a diagnostic context remains ambiguous; conceivably, its application could be tailored for focused reclassification endeavors in specific disease-associated genes.

The results show that, among CADD’s top 5 scoring variants, sSNV 9-139685876-G-A stands out as the sole variant predicted as likely pathogenic by SilVA. This variant has no clinical significance reported in literature yet, and is located on the transmembrane protein 141 (TMEM141), on chromosome 9. Although its function is not clearly known, it is speculated that TMEM141 might be involved in signal transduction across the membrane due to its loosely packed helices (Klammt et al., 2012). The other four variants that constitute CADD’s top 5 were deemed potentially pathogenic by SilVA: 19-7747163-G-T, 20-35807771-G-A, 3-73047268-G-A, 9-35809402-G-C. Despite none of them having clinical significance reported, all are located on disease associated genes (Table 1).

Another noteworthy result is that four of SilVA’s top 5 variants are encompassed within the 1% of variants predicted by CADD to be the most deleterious. All of them are located on genes which have been associated with disorders such as cancer, neurological and neurodegenerative diseases. Despite having no strong association with any disease according to MalaCards (Rappaport et al., 2013), The Negative Elongation Factor Complex Member E (NELF-E) was found to be involved in the regulation of HIV’s post infection transcription (Rao et al., 2006). It is also important to observe that none of SilVA’s top 5 scoring sSNVs have clinical significance reported in the literature yet.

The Ensemble Feature Selection analysis revealed that the most relevant feature for the effect prediction is “consequence”, denoting the variant’s impact on genomic features as provided by VEP. Notably, 25% of deleterious variants manifest beyond splicing sites, thus indicating that other mechanisms leading to deleteriousness are at play, such as impaired mRNA stability and codon usage bias. The mechanisms by which a synonymous variant can impair mRNA stability, thus, the efficiency of translation, are not readily discernible. Although there’s a possibility that it could impact translation initiation, it’s also feasible that sSNVs might induce a broader disruption in mRNP structure (Duan et al., 2003). When it comes to codon usage bias, it is directly associated with the optimization of fundamental cellular processes, such as speed and accuracy of translation (Hunt et al., 2014).

The remaining attributes within the top eight encompass conservation scores provided by various software tools. Conservation is one of the most important aspects for assessing the effect of any variant. Evolutionary processes have discarded most deleterious mutations, albeit variants of all levels of conservation are required for species adaptability (Miller et al., 2019). “Conservation” is also related to the non-random frequency of synonymous codon selection that varies from organism to organism, and even from gene to gene. The change of the preferred codon for an unusual one can influence the efficiency of gene expression processes and folding of the protein (Spencer et al., 2012). The significance of the proportion of heterochromatin state in 127 cell types, ranked ninth in our analysis, can be attributed to the nature of heterochromatin, primarily comprising inactive regions of the genome where gene expression is limited in most instances.

When categorizing the variants based on their allele frequency, it is important to consider the underlying assumption that detrimental variants generally exhibit low AF due to evolutionary processes (Buhr et al., 2016). Therefore, variant effect predictors lean to filter deleterious variants of higher frequency and neutral variants of lower frequency (Zeng and Bromberg, 2019). This contributes to the observation of a substantial percentage of variants from both classes, benign and deleterious, possessing an AF ≤ 0.1%, thereby being classified as super rare in this study.

The results of enrichment analyses for the 9635 genes where the super rare variants are situated indicate that metabolic pathways are significantly enriched in a substantial number of genes. However, it is noteworthy that these enriched pathways represent less than 35% of the total gene set associated with them. Although it is possible that many sSNVs may play a role in the diversity of metabolic pathways, it is also reasonable to assume that this is a very unspecific pathway, overrepresented in too many genes. The sSNV 6-152631823-C-T, one of the top-scoring SilVA variants, is located on the gene SYNE1, which has been associated with spinocerebellar ataxia. This pathway was found to be enriched in 66 genes. This analysis might reveal new targets for disease linkage due to its strength of association (FDR = 2.76x10−9).

Regarding the not rare deleterious variants, the findings reveal that all 39 sSNVs are situated within genes linked to one or more diseases, except for GAKP1. Interestingly, a missense variant in this gene was shared by three esophageal squamous cell carcinoma (ESCC) patients in a study that aimed to identify candidate susceptibility variants for ESCC (Donner et al., 2017). This gene encodes a protein that is highly similar to the mouse GKAP1 protein. In mice, GKAP1 is known to be mainly expressed in the testis, and its deletion increases sperm production (Wang et al., 2019).

The results for the positive selection analyses revealed that our best candidate for positive selection was found on the gene FOXD4L5, located on chromosome 9. Gene ontology annotations for the FOX family gene member predict it to enable transcription factor-DNA binding. FOXD4L5 was found to be one of the top 12 highly mutated genes in a familial lung cancer study (Kanwal et al., 2018), although further investigation is necessary to assess whether the mutations contribute to the development of cancer. Importantly, the sSNV that we found on FOXD4L5 was not one of those mutations and has never been associated with diseases to date.

The second PBS highest scoring variant overall is an A>G change and it’s located on the gene CHRFAM7A, on chromosome 15. This gene is the result of the duplication of exons 5 to 10 of CHRNA7 in fusion with FAM7A, a cluster of seven exons (A to F) located both upstream and downstream of CHRNA7 on chromosome 15 (Di Lascio et al., 2022). Although no clinical significance has yet been reported for this variant, CHRFAM7A has been widely associated with neurological disorders such as schizophrenia and bipolar disorder (Kunii et al., 2015). Further studies in leukocytes and macrophages revealed that CHRFAM7A plays an important role in the activation of the cholinergic anti-inflammatory pathway (Maroli et al., 2019).

The T>C variant is the third best candidate to be positively selected in the North-Western European population. No clinical significance has been reported to this variant. It is found on the gene HERC2, on chromosome 15, which belongs to the HERC gene family. This family is known to produce large proteins with multiple structural domains. HERC2 is related to ligase activity and ubiquitin protein ligase binding according to gene ontology annotations and was recently found to play an important role on the regulation of nucleolar localization of the helicases (Zhu et al., 2021). HERC2 is frequently downregulated in numerous types of cancers due to its critical role on chromosomal stability (Wu et al., 2018), and variants in this gene have been associated with developmental delay (Puffenberger et al., 2012).

The fourth best candidate to be positively selected on the North-Western European population is a T>C variant found on the gene NOMO3, on chromosome 16. It codes for a transmembrane protein that is highly conserved among human tissues (Sun et al., 2022) and participates in a complex that takes part in the Nodal signaling pathway during vertebrate development (Haffner et al., 2004). A study of a five generation family with Multiple Synostoses Syndrome Type 4 presenting reduction of the GDF6 gene expression reported NOMO3 as severely downregulated, which might indicate that it plays a role in the GDF6 pathway to skeletal joint development and ossification (Clarke et al., 2021). An immune profiling of Medullary Thyroid Cancer (MTC) reported NOMO3 to be highly expressed in MTC, configuring as potential tumor-associated antigen (Pozdeyev et al., 2020). Importantly, the sSNV discussed here has no clinical significance reported in the literature.

The G>A sSNV found on the gene LRRC37A has no clinical significance reported. The LRRC37 gene family is located on a complex region on chromosome 17 subject to high linkage disequilibrium (Bekpen et al., 2012). Several studies reported that genes located in this region, such as LRRC37, are associated with Parkinson Disease risk (Bowles et al., 2022). Furthermore, Bowles et al. (2022) demonstrated that LRRC37A produces a membrane-associated protein that contributes in chemotaxis, astroglial inflammation and cellular migration.

The aforementioned sSNVs exhibited positive selection in the north-western European population. Notably, the overall third highest scoring sSNV demonstrated positive selection in the East Asian population. This particular variant involves a G>T change and is situated on the gene NPIPB5, which is located on chromosome 16. No clinical significance has been reported to this variant, but a recent study described the gene NPIPB5 as a putative novel prognostic biomarker for clear cell renal cell carcinoma (Wang et al., 2022).

On the gene WASHC2A, we find the T>G variant, a good candidate for positive selection on the East Asian population, with no association with diseases to date. The WASHC2A gene, located on chromosome 10, is part of the gene family FAM21 which is the largest component of the multiprotein complex called The Wiskott-Aldrich syndrome protein and SCAR homologue (WASH) complex (Lee et al., 2016). This complex takes part on the endosomal sorting pathway, where endosomes sort the proteins for lysosome degradation or the recycling pathway (Bonifacino and Rojas, 2006). Depletion of the WASH complex results in endosomal sorting defects in subsequent pathways (Gomez and Billadeau, 2009). Furthermore, Vincendeau et al. (2010) demonstrated that a protein fragment which interacts with the HIV regulatory protein Rev, can control HIV replication and is located in the highly conserved proteins encoded by FAM21 genes.

Another noteworthy variant is the T>C change, located on the gene TPSB2, on chromosome 16. β tryptases are tetrameric serine proteases secreted by mast cells upon activation (Schwartz and Irani, 2000). Elevated serum level of mature Tryptase β serves as diagnostics for mastocytosis (Valent et al., 2001).

The fourth sSNV most likely to be positively selected in the East Asian population is a T>C change on the gene PKD1L2, located on chromosome 16. This gene encodes a protein member of the polycystin family, which is composed of membrane proteins that function as ion-channel regulators and share significant homology to each other (Yuasa et al., 2004). More specifically, the product of PKD1L2 has several alternative splicing forms and binds to specific G-protein subunits, which are responsible for transducing extracellular signals into the cell (Yuasa et al., 2004). Besides being associated with polycystic kidney disease, a study in the Korean population revealed that a copy number variation in PKD1L2 is related with colorectal cancer predisposition (Park et al., 2017). Additionally, PKD1L2 features a four-mRNA model recently constructed for prediction of breast cancer prognosis (Qi et al., 2019).

Finally, the T>C variant is our last candidate for positive selection on the East Asian population. It is situated on the gene RGPD3, a member of the RGP gene family, located in the cluster of Ran-binding protein-related genes, on chromosome 2. This gene family is composed of eight partial copies of the RanBP2 gene, originating from intrachromosomal segmental duplication, with the posterior acquisition of the GRIP domain. It is likely that RGP genes play a role in intracellular trafficking (Ciccarelli et al., 2005). Moreover, a study on what influences craniofacial morphology on Northern Han Chinese suggested that the RGPD3 gene is associated with the morphology of the nose and ears (Wu et al., 2019).

It is significant to point out that all positively selected sSNV here described have no clinical significance reported in the literature. Most are located on functionally important, pleiotropic genes that may be benefiting from more frequently used codons that confer better pace of synthesis and cotranslational folding of proteins. This fitness benefit would explain why these variants are likely being positively selected in, at least, two populations. However, accurately determining the specific advantages conferred by such positive selection poses a significant challenge due to the prevailing focus of most studies on diseases linked to specific genetic variations, genes or genomic regions. Further investigation taking into account whether causal sSNVs are also subject to evolutionary constraints and/or associated with life-history traits of a population, as well as with diseases, may complement these findings.

The results regarding positive selection must be taken with caution. Most methods used in population genetics were developed considering natural populations, that is, geographically dispersed populations. Here, we have to take into account that gnomAD uses ancestry informative markers (AIM) for the biogeographical classification of populations by training a random forest model using samples with known ancestry (Karczewski et al., 2020). Grouping individuals based on shared characteristics in order to make sense of the biological variation is a complex task and has many possibilities depending on the population concept being used. The outlier analysis can also be affected by the different number of samples in each population (see Materials and Methods) and by the lack of genotype level information on the gnomAD files. We were not able to use data from the 1000 Genomes Project to access genotype information in this analysis as roughly only 20% of the sSNVs present in the gnomAD data were present at the 1000 Genomes database, although it comprises most of gnomAD samples for which exome sequencing is available.

In this study, two sophisticated predictors have been employed, both of which incorporate a range of features and mechanisms to comprehensively assess the potential impact of sSNVs on the human exome. The resulting framework provides a highly detailed and accurate prediction of the effect of these variants on gene function. Vihinen (2022) suggested ”unsense” as a new nomenclature for synonymous variants having an effect on the protein. This may fundamentally change the way in which predictors annotate these variants in the future. Therefore, accurately annotating the deleteriousness of synonymous variants is increasingly challenging, and requires a more comprehensive approach that takes into account the diverse mechanisms by which these variants can affect gene function.

Conclusions

In this study we have shown that a significant number of sSNVs can have an impact on protein function and propose a framework for sSNVs variant prioritization. This effect seems to be related mainly to alteration of splice sites and loss of preferred codons, suggesting an effect on protein folding. The fact that a substantial portion of the predicted deleterious sSNVs have ultra rare allelic frequency in the population and are present in disease-related genes strengthens our predictions. Finally, outlier analyses revealed that at least ten benign sSNVs are likely to be under positive selection in two populations. Taken together these results give voice to the so-called silent mutations and caution that these variants must be included in variant prioritization analysis. As with any prediction study, further experimental confirmation should be advised.

Acknowledgements

The authors would like to thank Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) for the fellowship provided to ACM (process no. 88887.369793/2019-00), and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) for the fellowship provided to UM (Process no. 307038/2015-7). This research was supported by Fundo de Incentivo à Pesquisa e Eventos (FIPE) at Hospital de Clínicas de Porto Alegre (HCPA) and by the Graduate Program in Genetics and Molecular Biology (PPGBM) at Universidade Federal do Rio Grande do Sul (UFRGS).

References

• Bekpen C, Tastekin I, Siswara P, Akdis CA and Eichler EE (2012) Primate segmental duplication creates novel promoters for the LRRC37 gene family within the 17q21.31 inversion polymorphism region. Genome Res 22:1050-1058.
• Bommert A, Sun X, Bischl B, Rahnenführer J and Lang M (2020) Benchmark for filter methods for feature selection in high-dimensional classification data. Compu Stat Data Anal 143:106839.
• Bonifacino JS and Rojas R (2006) Retrograde transport from endosomes to the trans-Golgi network. Nat Rev Mol Cell Biol 7:568-579.
• Bonin S, Donada M, Bussolati G, Nardon E, Annaratone L, Pichler M, Chiaravalli AM, Capella C, Hoefler G and Stanta G (2016) A synonymous EGFR polymorphism predicting responsiveness to anti-EGFR therapy in metastatic colorectal cancer patients. Tumour Biol 37:7295-7303.
• Bowles KR, Pugh DA, Liu Y, Patel T, Renton AE, Bandres-Ciga S, Gan-Or Z, Heutink P, Siitonen A, Bertelsen S et al (2022) 17q21.31 sub-haplotypes underlying H1-associated risk for Parkinson’s disease are associated with LRRC37A/2 expression in astrocytes. Mol Neurodegener 17:48.
• Buhr F, Jha S, Thommen M, Mittelstaet J, Kutz F, Schwalbe H, Rodnina MV and Komar AA (2016) Synonymous codons direct cotranslational folding toward different protein conformations. Mol Cell 61:341-351.
• Buske OJ, Manickaraj A, Mital S, Ray PN and Brudno M (2015) Identification of deleterious synonymous variants in human genomes. Bioinformatics 31:799.
• Cartegni L, Chew SL and Krainer AR (2002) Listening to silence and understanding nonsense: Exonic mutations that affect splicing. Nat Rev Genet 3:285-298.
• Chen R, Davydov EV, Sirota M and Butte AJ (2010) Non-synonymous and synonymous coding SNPs show similar likelihood and effect size of human disease association. PLoS One 5:e13574.
• Ciccarelli FD, Von Mering C, Suyama M, Harrington ED, Izaurralde E and Bork P (2005) Complex genomic rearrangements lead to novel primate gene function. Genome Res 15:343-351.
• Clarke RA, Fang Z, Murrell D, Sheriff T and Eapen V (2021) Knockdown in a family with multiple synostosis syndrome and speech impairment. Genes (Basel) 12:1354.
• Cooper GM, Goode DL, Ng SB, Sidow A, Bamshad MJ, Shendure J and Nickerson DA (2010) Single-nucleotide evolutionary constraint scores highlight disease-causing mutations. Nat Methods 7:250-251.
• Di Lascio S, Fornasari D and Benfante R (2022) The human-restricted Isoform of the α7 nAChR, CHRFAM7A: A double-edged sword in neurological and inflammatory disorders. Int J Mol Sci 23:3912.
• Diederichs S, Bartsch L, Berkmann JC, Fröse K, Heitmann J, Hoppe C, Iggena D, Jazmati D, Karschnia P, Linsenmeier M et al (2016) The dark matter of the cancer genome: Aberrations in regulatory elements, untranslated regions, splice sites, non-coding RNA and synonymous mutations. EMBO Mol Med 8:442-457.
• Donner I, Katainen R, Tanskanen T, Kaasinen E, Aavikko M, Ovaska K, Artama M, Pukkala E and Aaltonen LA (2017) Candidate susceptibility variants for esophageal squamous cell carcinoma. Genes Chromosomes Cancer 56:453-459.
• Duan J, Wainwright MS, Comeron JM, Saitou N, Sanders AR, Gelernter J and Gejman PV (2003) Synonymous mutations in the human dopamine receptor D2 (DRD2) affect mRNA stability and synthesis of the receptor. Hum Mol Genet 12:205-216.
• Foll M and Gaggiotti O (2008) A genome-scan method to identify selected loci appropriate for both dominant and codominant markers: A Bayesian perspective. Genetics 180:977-993.
• Gartner JJ, Parker SC, Prickett TD, Dutton-Regester K, Stitzel ML, Lin JC, Davis S, Simhadri VL, Jha S, Katagiri N et al (2013) Whole-genome sequencing identifies a recurrent functional synonymous mutation in melanoma. Proc Natl Acad Sci U S A 110:13481-13486.
• Gomez TS and Billadeau DD (2009) A FAM21-containing WASH complex regulates retromer-dependent sorting. Dev Cell 17:699-711.
• Haffner C, Frauli M, Topp S, Irmler M, Hofmann K, Regula JT, Bally-Cuif L and Haass C (2004) Nicalin and its binding partner Nomo are novel Nodal signaling antagonists. EMBO J 23:3041-3050.
• Holsinger KE and Weir BS (2009) Genetics in geographically structured populations: Defining, estimating and interpreting F(ST). Nat Rev Genet 10:639-650.
• Holte RC (1993) Very simple classification rules perform well on most commonly used datasets. Mach Learn 11:63-90.
• Hunt RC, Simhadri VL, Iandoli M, Sauna ZE and Kimchi-Sarfaty C (2014) Exposing synonymous mutations. Trends Genet 30:308-321.
• Kanwal M, Ding XJ, Ma ZH, Li LW, Wang P, Chen Y, Huang YC and Cao Y (2018) Characterization of germline mutations in familial lung cancer from the Chinese population. Gene 641:94-104.
• Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, Collins RL, Laricchia KM, Ganna A, Birnbaum DP et al (2020) The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581:434-443.
• Klammt C, Maslennikov I, Bayrhuber M, Eichmann C, Vajpai N, Chiu EJ, Blain KY, Esquivies L, Kwon JH, Balana B et al (2012) Facile backbone structure determination of human membrane proteins by NMR spectroscopy. Nat Methods 9:834-839.
• Kunii Y, Zhang W, Xu Q, Hyde TM, McFadden W, Shin JH, Deep-Soboslay A, Ye T, Li C, Kleinman JE et al (2015) CHRNA7 and CHRFAM7A mRNAs: Co-localized and their expression levels altered in the postmortem dorsolateral prefrontal cortex in major psychiatric disorders. Am J Psychiatry 172:1122-1130.
• Lee S, Chang J and Blackstone C (2016) FAM21 directs SNX27-retromer cargoes to the plasma membrane by preventing transport to the Golgi apparatus. Nat Commun 7:10939.
• Maroli A, Di Lascio S, Drufuca L, Cardani S, Setten E, Locati M, Fornasari D and Benfante R (2019) Effect of donepezil on the expression and responsiveness to LPS of CHRNA7 and CHRFAM7A in macrophages: A possible link to the cholinergic anti-inflammatory pathway. J Neuroimmunol 332:155-166.
• Miller M, Vitale D, Kahn PC, Rost B and Bromberg Y (2019) FUNTRP: Identifying protein positions for variation driven functional tuning. Nucleic Acids Res 47:e142.
• Nackley AG, Shabalina SA, Tchivileva IE, Satterfield K, Korchynskyi O, Makarov SS, Maixner W and Diatchenko L (2006) Human catechol-O-methyltransferase haplotypes modulate protein expression by altering mRNA secondary structure. Science 314:1930-1933.
• Narum SR and Hess JE (2011) Comparison of FST outlier tests for SNP loci under selection. Mol Ecol Resour 11:184-194.
• Palagano E, Susani L, Menale C, Ramenghi U, Berger M, Uva P, Oppo M, Vezzoni P, Villa A and Sobacchi C (2017) Synonymous mutations add a layer of complexity in the diagnosis of human osteopetrosis. J Bone Miner Res 32:99-105.
• Park C, Kim JI, Hong SN, Jung HM, Kim TJ, Lee S, Kim SJ, Kim HC, Kim DH, Cho B et al (2017) A copy number variation in PKD1L2 is associated with colorectal cancer predisposition in Korean population. Int J Cancer 140:86-94.
• Pozdeyev N, Erickson TA, Zhang L, Ellison K, Rivard CJ, Sams S, Hirsch FR, Haugen BR and French JD (2020) Comprehensive immune profiling of medullary thyroid cancer. Thyroid 30:1263-1279.
• Puffenberger EG, Jinks RN, Wang H, Xin B, Fiorentini C, Sherman EA, Degrazio D, Shaw C, Sougnez C, Cibulskis K et al (2012) A homozygous missense mutation in HERC2 associated with global developmental delay and autism spectrum disorder. Hum Mutat 33:1639-1646.
• Qi L, Yao Y, Zhang T, Feng F, Zhou C, Xu X and Sun C (2019) A four-mRNA model to improve the prediction of breast cancer prognosis. Gene 721:144100.
• Quinlan JR (1986) Induction of decision trees. Mach Learn 1:81-106.
• Rao JN, Neumann L, Wenzel S, Schweimer K, Rosch P and Wohrl BM (2006) Structural studies on the RNA-recognition motif of NELF E, a cellular negative transcription elongation factor involved in the regulation of HIV transcription. Biochem J 400:449-456.
• Rappaport N, Nativ N, Stelzer G, Twik M, Guan-Golan Y, Stein TI, Bahir I, Belinky F, Morrey CP, Safran M et al (2013) MalaCards: An integrated compendium for diseases and their annotation. Database (Oxford) 2013:bat018.
• Rentzsch P, Witten D, Cooper GM, Shendure J and Kircher M (2019) CADD: Predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res 47:D886-D894.
• Richards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, Grody WW, Hegde M, Lyon E, Spector E et al (2015) Standards and guidelines for the interpretation of sequence variants: A joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med 17:405-424.
• Schwartz LB and Irani AM (2000) Serum tryptase and the laboratory diagnosis of systemic mastocytosis. Hematol Oncol Clin North Am 14:641-657.
• Shen X, Song S, Li C and Zhang J (2022) Synonymous mutations in representative yeast genes are mostly strongly non-neutral. Nature 606:725-731.
• Spencer PS, Siller E, Anderson JF and Barral JM (2012) Silent substitutions predictably alter translation elongation rates and protein folding efficiencies. J Mol Biol 422:328-335.
• Sun Y, Li T and Qian X (2022) Biological role of nodal modulator: A comprehensive review of the last two decades. DNA Cell Biol 41:336-341.
• Tigano A, Shultz AJ, Edwards SV, Robertson GJ and Friesen VL (2017) Outlier analyses to test for local adaptation to breeding grounds in a migratory arctic seabird. Ecol Evol 7:2370-2381.
• Valent P, Horny HP, Escribano L, Longley BJ, Li CY, Schwartz LB, Marone G, Nunez R, Akin C, Sotlar K et al (2001) Diagnostic criteria and classification of mastocytosis: A consensus proposal. Leuk Res 25:603-625.
• Vihinen M (2022) When a synonymous variant is nonsynonymous. Genes (Basel) 13:1485.
• Vincendeau M, Kramer S, Hadian K, Rothenaigner I, Bell J, Hauck SM, Bickel C, Nagel D, Kremmer E, Werner T et al (2010) Control of HIV replication in astrocytes by a family of highly conserved host proteins with a common Rev-interacting domain (Risp). AIDS 24:2433-2442.
• Wang H, Wang G, Dai Y, Li Z, Zhu Y and Sun F (2019) Functional role of GKAP1 in the regulation of male germ cell spontaneous apoptosis and sperm number. Mol Reprod Dev 86:1199-1209.
• Wang J, Huang F, Zhao J, Huang P, Tan J, Huang M, Ma R, Xiao Y, He S, Wang Z et al (2022) Tumor-infiltrated CD8+ T Cell 10-gene signature related to clear cell renal cell carcinoma prognosis. Front Immunol 13:930921.
• Wright S (1951) The genetical structure of populations. Ann Eugen 15:323-354.
• Wu W, Rokutanda N, Takeuchi J, Lai Y, Maruyama R, Togashi Y, Nishikawa H, Arai N, Miyoshi Y, Suzuki N et al (2018) HERC2 facilitates BLM and WRN helicase complex interaction with RPA to suppress G-quadruplex DNA. Cancer Res 78:6371-6385.
• Wu W, Zhai G, Xu Z, Hou B, Liu D, Liu T, Liu W and Ren F (2019) Whole-exome sequencing identified four loci influencing craniofacial morphology in northern Han Chinese. Hum Genet 138:601-611.
• Yi X, Liang Y, Huerta-Sanchez E, Jin X, Cuo ZX, Pool JE, Xu X, Jiang H, Vinckenbosch N, Korneliussen TS et al (2010) Sequencing of 50 human exomes reveals adaptation to high altitude. Science 329:75-78.
• Yu G, Wang LG, Han Y and He QY (2012) ClusterProfiler: An R package for comparing biological themes among gene clusters. OMICS 16:284-287.
• Yuasa T, Takakura A, Denker BM, Venugopal B and Zhou J (2004) Polycystin-1L2 is a novel G-protein-binding protein. Genomics 84:126-138.
• Zeng Z and Bromberg Y (2019) Predicting functional effects of synonymous variants: A systematic review and perspectives. Front Genet 10:914.
• Zhao S, Guo Y and Shyr Y (2020) KEGGprofile: An annotation and visualization package for multi-types and multi-groups expression data in KEGG pathway. R package version 1.32.0., https://bioconductor.org/packages/3.12/bioc/html/KEGGprofile.html
  » https://bioconductor.org/packages/3.12/bioc/html/KEGGprofile.html
• Zhu M, Wu W, Togashi Y, Liang W, Miyoshi Y and Ohta T (2021) HERC2 inactivation abrogates nucleolar localization of RecQ helicases BLM and WRN. Sci Rep 11:360.

Internet Resources

• Romanski P, Kotthoff L and Schratz P (2021) FSelector: Selecting attributes, R package version 0.33, 33, https://CRAN.R-project.org/package=FSelector (accessed 28 october 2020).
  » https://CRAN.R-project.org/package=FSelector

• Data Availability
  All data used in this work are publicly available at the Genome Aggregation Database (https://gnomad.broadinstitute.org/).

Supplementary material

The following online material is available for this article:

Figure S1 - Dataset schematic view. Proportion of synonymous variants compared to the initial downloaded dataset, and proportion of synonymous variants in both classes (benign and deleterious).

Figure S2 - Classification of the variants according to the allele frequency.

Figure S3 - Manhattan plots showing the results for the Bayescan analysis.

Figure S4 - Gene Ontology (GO) enrichment analysis of the genes in which the super rare deleterious variants are located.

Figure S5 - Gene Ontology (GO) enrichment analysis of the genes in which the rare deleterious variants are located.

Table S1 - Ranking of CADD relevant features from the Ensemble Feature Selection analysis and the description of the features (Rentzsch , 2019).

Table S2 - Description of CADD features removed from the Ensemble Feature Selection analysis.

Table S3 - Name description of diseases from Table 2.

File S1 - Tab-separated file (.txt) containing all sSNVs classified as benign in this study, as well as its relevant features and scores.

File S2 - Tab-separated file (.txt) containing all sSNVs classified as deleterious in this study, as well as its relevant features and scores.

File S3 - Comma-separated file (.csv) showing the results for the KEGG analysis on the super rare deleterious sSNVs.

File S4 - Comma-separated file (.csv) showing the results for the Gene Onothology analysis on the super rare deleterious sSNVs.

File S5 - Comma-separated file (.csv) showing the results for the KEGG analysis on the rare deleterious sSNVs.

File S6 - Comma-separated file (.csv) showing the results for the Gene Onthology analysis on the rare deleterious sSNVs.

Edited by

• Associate Editor
  Lavínia Schüler-Faccini

Publication Dates

• Publication in this collection
  22 Jan 2024
• Date of issue
  2023

History

• Received
  03 May 2023
• Accepted
  10 Dec 2023