Link between short tandem repeats and translation initiation site selection

Despite their vast biological implication, the relevance of short tandem repeats (STRs)/microsatellites to the protein-coding gene translation initiation sites (TISs) remains largely unknown. We performed an Ensembl-based comparative genomics study of all annotated orthologous TIS-flanking sequences in human and 46 other species across vertebrates, on the genomic DNA and cDNA platforms (755,956 TISs), aimed at identifying human-specific STRs in this interval. The collected data were used to examine the hypothesis of a link between STRs and TISs. BLAST was used to compare the initial five amino acids (excluding the initial methionine), codons of which were flanked by STRs in human, with the initial five amino acids of all annotated proteins for the orthologous genes in other vertebrates (total of 5,314,979 pair-wise TIS comparisons on the genomic DNA and cDNA platforms) in order to compare the number of events in which human-specific and non-specific STRs occurred with homologous and non-homologous TISs (i.e., ≥ 50% and < 50% similarity of the five amino acids). We detected differential distribution of the human-specific STRs in comparison to the overall distribution of STRs on the genomic DNA and cDNA platforms (Mann Whitney U test p = 1.4 × 10−11 and p < 7.9 × 10−11, respectively). We also found excess occurrence of non-homologous TISs with human-specific STRs and excess occurrence of homologous TISs with non-specific STRs on both platforms (p < 0.00001). We propose a link between STRs and TIS selection, based on the differential co-occurrence rate of human-specific STRs with non-homologous TISs and non-specific STRs with homologous TISs.


Introduction
An increasing number of human protein-coding genes are unraveled to consist of alternative translation initiation sites (TISs), which are selected based on complex and yet not fully known scanning mechanisms [1,2]. The alternative TISs result in various protein structures and functions [3,4]. Selection of TISs and the level of translation and protein synthesis depend partially on the cis-regulatory elements in the mRNA sequence and its secondary structure such as the formation of hair-pins and thermal stability [5][6][7]. Genomic DNA cis-elements can also affect translation and TISs through various mechanisms (for a review see [8]).
One of the important and understudied cis-regulatory elements affecting translation is short tandem repeats (STRs)/microsatellites. In physiological terms, STRs can dramatically influence TIS and the amount of protein synthesis. Poly(A) tracts in the 5′-untranslated region (UTR) are important sites for translation regulation in yeast. These poly(A) tracts can interact with translation initiation factors or poly(A) binding proteins (PABP) to either increase or decrease translation efficiency. Pre-AUG A N can enhance internal ribosomal entry both in the presence of PABP and eIF-4G in Saccharomyces cerevisiae [9], and in the complete absence of PABP and eIF-4G [10]. Biased distribution of dinucleotide repeats is a known phenomenon in the region immediately upstream of the TISs in Escherichia coli [11]. In pathological instances, expansion of STRs in the RNA structure results in toxic RNAs and non-AUG translation and the development of several human-specific neurological [12][13][14].
This research aimed to examine a possible link between STRs and TIS selection through studying the occurrence rate of TIS-flanking human-specific and non-specific STRs with homologous and non-homologous proteins.
For each gene in each species, its Ensembl ID, all the annotated transcript IDs, the genomic DNA sequence, the cDNA, and the coding DNA sequence (CDS) were retrieved (the list of genes is available upon request). The genomic DNA, CDS, and the annotated cDNAs were downloaded using REST API from the Ensembl database. The first start codon for each transcript was determined using BLAST between the CDS and cDNA. The 120 bp genomic DNA and cDNA interval upstream of the start codon (ATG) were investigated for the presence of STRs of ≥ 3-repeats (Additional file 1).

Retrieval of gene IDs across species
Using the Enhanced REST API tools, a set of functions were developed to analyze genes and their transcripts information, including func_get_ensemblID and func_ get_TranscriptsID. The genomic DNA, cDNA, and CDS sequences of genes and their respective transcripts were obtained using func_get_GenomicSequence, func_get_cD NASequence, and func_get_CDSSequence functions.
Identification of STRs in the human TIS-flanking genomic and cDNA intervals A general method of finding human-specific and nonspecific STRs (≥ 3-repeats in all classes of STRs, except the mononucleotide repeats, in which STRs of ≥ 6-repeats were studied) for each individual gene was developed and applied as follows: the 120 bp genomic DNA and cDNA sequence upstream of the TISs of all annotated protein-coding gene transcripts was screened in human and 46 other species across vertebrates for the presence of STRs. A list of all STRs and their abundance was prepared for each gene in every species. The data obtained on the human STRs was compared to those of other species, and the term "human-specific" was applied to STRs that were not detected at ≥ 3-repeats in any other species. Exceptionally, in the mononucleotide category, the threshold of repeats for "human-specificity" was set at > 6-repeats. The relevant pseudo-code for the identification of repeated substrings was used for STR identification (Additional file 2).
Mann-Whitney U test was used to compare the distribution of human-specific vs. the overall (specific and non-specific) STR distribution in human.

TIS homology threshold estimation
Weighted homology scoring was performed, in which the initial five amino acids (excluding the initial methionine) of the human protein-coding TISs, codons of which were flanked by STRs, were BLASTed (compared using BLAST) against all initial protein-coding five amino acids annotated for the orthologous genes in 46 species across vertebrates [(3,872,779 pair-wise TIS comparisons on the genomic DNA platform (Ensembl 91) and 1,442,200 pair-wise TIS comparisons on the cDNA platform (Ensembl 92)]. The above was aimed at comparing the number of events in which human-specific and non-specific STRs occurred with homologous and non-homologous TISs.
The following equation was developed for the weighted scoring of homology (Eq. 1), where A refers to the five amino acid sequence (excluding the initial methionine M), codons of which were flanked by a STR at the genomic DNA or cDNA sequence, j refers to the gene, and B refers to all the transcripts of the same gene that contain the STR in other species.
If M is the first methionine amino acid of two sequences, for all five successive positions represented by i in the equation, we defined five weight coefficients W 1 to W 5 based on the importance of the amino acid position, i.e., proximity to the methionine starting codon, observed in the W vector. The degree of homology between the two sequences A and B was calculated using function φ for all five positions with the operations P L¼5 i¼1 W i φðA jik ; B jik 0 Þ . We repeated this operation for k transcripts, where k stands for the number of transcripts in human. k ′ refers to all transcripts of the gene j in other species.
; for all k and k 0 ð1Þ Homology of the five amino acids, and therefore the TISs, was inferred based on the %similarity scoring. We validated the homology threshold by measuring the %similarity of 3000 random pairs of human proteins (the first five amino acids excluding the initial methionine), where similarity of ≥ 50% was virtually non-existent in that sample (6 in 3000, false positive rate = 0.001) (Additional file 3).
Finally, the two by two table and Fisher exact statistics were used to examine the link between STRs and TISs.

Results
Genome-scale distribution of human STRs in the 120 bp upstream sequence of TISs Genomic DNA platform Mono-and dinucleotide STRs dominated STRs of > 1000 counts, and the (T)6 mononucleotide repeat was the most abundant STR in this interval, succeeded by the (CT)3 and (TC)3 dinucleotide STRs (Fig. 1). Trinucleotide STRs were less abundant, observed at counts between 100 and 1000, and predominated by GC-rich composition such as (GGC)3, (GCC)3, (CCG)3, and (GCG)3. In the non-GC composition, (CTC)3 and (CCT)3 were the most common trinucleotide STRs. Tetra-, penta-, and hexanucleotide STRs were at lesser abundance than the above categories and observed at < 100 counts, where (CCTC)3 and (CCGC)3 were the most abundant tetranucleotide STRs. Only three pentanucleotide STR classes, (GGGGC)3, (TTTTG)3, and (GTTTT)3, were observed at counts > 10 in the screened interval. The overall distribution of STRs in the 120 bp TISflanking cDNA sequences was significantly different in comparison to the genomic DNA STRs (Fig. 2). In comparison to the genomic DNA platform on which T(6) was the most abundant STR, GC-rich dinucleotide repeats were the most abundant on the cDNA platform. Numerous other instances at high, medium, and low abundance differentiated the genomic DNA vs. cDNA platforms (e.g., differential abundance of (T)8, (GT)3, (TA)3, and (CA)3 between the two platforms).
Human-specific STR fingerprints on the TIS-flanking genomic DNA and cDNA platforms and differential distribution of these compartments in comparison to the overall STR distribution Genomic DNA platform The flanking sequence of 755,956 TISs was screened in human and 46 other species in order to identify human-specific STRs. One thousand eight hundred eighty-seven genes contained human-specific TIS-flanking STRs on the genomic DNA platform, which were of a wide range of nucleotide compositions of mono-, di-, tri-, tetra-, penta-, and hexanucleotide repeats, of which poly(A) and poly(T) STRs were the longest (the 1st percentile, based on STR length, is listed in Table 1, and the list of all genes is provided as Additional file 4).
As an extreme example, the TIS of the NVL gene was flanked by a human-specific (T)22 STR, which was the longest STR detected in a human protein-coding gene TIS-flanking sequence. The TIS of the gene, SULT1A3, was flanked by the longest poly(A) at (A)18. Short-and medium-length STRs were also detected in the humanspecific compartment (Additional file 4).
Significant skewing was observed in the distribution of human-specific STRs (Fig. 3) vs. the overall (i.e., human-specific and non-specific) STRs (Fig. 1) (Mann Whitney U test, p = 1 × 10 −5 ). While the (GC)3 and (CG)3 dinucleotide STRs were enriched in the overall STR compartment, their abundance was significantly lower in the human-specific compartment. Instead, (CA)3 and (AC)3 were significantly more abundant in the human-specific compartment. Differences in the distribution of tri-and tetranucleotide STRs were also observed between the two compartments. While trinucleotide and tetranucleotide STRs of GC composition were more abundant in the overall compartment, non-GC STR compositions (e.g., GGA, TTC, GCA, and ATAA) were more abundant in the human-specific compartment.

cDNA platform
Two thousand six hundred genes contained human-specific STRs in their TIS cDNA flanking sequence (the 1st percentile based on length is represented in Table 2 and The distribution of human-specific STRs on the cDNA platform (Fig. 4) was unique to this platform and different from the overall STR distribution on the cDNA platform (Fig. 2) (Mann-Whitney U test p = 1 × 10 −5 ). While the CG and CG dinucleotide STRs were more abundant in the overall distribution, CT, CA, and TG repeats were more abundant in the human-specific compartment. Various other differences were detected in other classes of STRs (e.g., more abundance of non-GC compositions in the trinucleotide STRs such as GGA, GCT, and CTG).

STRs link to TIS selection on both genomic DNA and cDNA platforms
We examined the hypothesis that there may be a link between STRs and TIS selection. The initial five amino acids (excluding the initial methionine) of the human protein sequences, codons of which were flanked by STRs at the genomic DNA and cDNA, were BLASTed (compared using BLAST) against the initial five amino acids of all the proteins annotated for the orthologous genes in 46 species across vertebrates in order to compare the number of events in which human-specific and non-specific STRs occurred with homologous and non-homologous TISs (≥ 50% and < 50% similarity of the five amino acids). Total of 5,314,979 pair-wise TIS comparisons were performed, and significant correlation was observed between STRs and TIS selection both on the genomic DNA (Fig. 5a) and cDNA platforms (Fig. 5b) (p < 0.00001), where there was excess occurrence of non-homologous TISs with human-specific STRs, and vice versa (i.e., excess occurrence of homologous TISs with non-specific STRs).

Discussion
In this study, we characterized the genome-scale STR landscape of the immediate 120 bp upstream sequence of human TISs on the genomic DNA and cDNA platforms, cataloged the human-specific compartment of these STRs, and investigated a possible link between STRs and TIS selection. Our findings provide the first evidence of a link between STRs and TIS selection on both platforms. This link is primarily deduced based on the differential co-occurrence rate of human-specific STRs with non-homologous TISs and non-specific STRs with homologous TISs. Sequence similarity searches can reliably identify "homologous" proteins or genes by detecting excess similarity [40]. The TIS homology threshold of ≥ 50% was validated based on 3000 random similarity scorings of the initial protein-coding five amino acids (excluding the initiating methionine) of human proteins, in which that threshold was non-existent in effect (false-positive rate = 0.001). This scoring methodology was consistently applied to the TISs linked to human-specific and non-specific STRs.
We also observed differential distribution of the human-specific STRs vs. the overall distribution of STRs on both genomic and cDNA platforms. Importantly, each platform had a unique pattern of STR distribution, indicating differential selection of STRs based on their location and evolutionary course. Genome-scale skewing of STRs, albeit at a lesser scale of STR classes, was reported by our group in a preliminary study of the gene core promoter interval [36].
It is imperative to envision that human-specific cis elements at the mRNA and DNA may result in the production of proteins that are specific to humans. The RNA structure influences recruitment of various RNA binding Fig. 3 Distribution of the human-specific STRs in the TIS-flanking genomic DNA sequence. A significant skewing was observed between this compartment and the compartment containing the overall (human-specific and non-specific) STRs. The abundance of STRs is sorted in the ascending order proteins and determines alternative TISs [41]. Indeed, the ribosomal machinery has the potential to scan and use several open reading frames (ORFs) at a particular mRNA species [42]. When located at the 5′ or 3′ UTR, STRs can modulate translation, the effect of which has biological and pathological implications [13,43,44]. The disorders linked to the 5′ UTR STRs encompass a number of human-specific neurological disorders.
On the genomic DNA platform, proximity to the splice sites may increase the biological/pathological implication of repeats [45]. Similar to the cDNA STRs, we observed significant enrichment of non-homologous TISs co-occurring with human-specific genomic STRs, which were substantially near the exons (within 120 bp upstream of the TISs).
Gene Ontology (GO) search yielded a variety of terms across the identified genes, including neuron cell fate specification, multicellular organism development, translation initiation factor activity, and cell differentiation (https://www.ebi.ac.uk/QuickGO/), examples of which are presented in Tables 1 and 2. EMBOSS Needle (https://www.ebi.ac.uk/Tools/psa/ emboss_needle) pair-wise comparison was performed between human and three other species (chimpanzee, macaque, and mouse), of the proteins encoded by the transcripts in Table 1 (Fig. 6a), Table 2 (Fig. 6b), and several randomly selected proteins, codons of which were  flanked by non-specific STRs (Fig. 6c). Identity scores were considerably lower across the three species for the genes in Tables 1 and 2, compared to the identity scores for the genes in the non-specific STR compartment. A number of the identified genes such as ACAT1 (Table 1) and SGIP1 (Table 2) confer risk for diseases or endophenotypes that are predominantly specific to the human species, such as complex psychiatric disorders [46,47]. SULT1A4 (Table 1) plays a critical role in neurotransmitter metabolism in the human brain and is also linked to neurodegeneration [48]. APOA2 (Additional file 4) along with several other lipoproteins is linked to cognitive health [49]. In another remarkable example, TBR1 (Additional file 5) is involved in FOXP2 gene expression, which has pivotal role in speech and language in human [50]. SRGAP2 (Additional file 5) family proteins may have increased the density of dendritic spines and promoted neoteny of the human brain during crucial periods of human evolution [51].
GO terms, protein conservation comparisons, and phenotypes stated above are only a few examples of the identified genes, in which human-specific STRs and the linked TISs may contribute to human evolution and disease. Future studies are warranted to examine the implication of the identified STRs and genes at the inter-and intra-species levels.

Conclusion
We characterized the landscape of STRs at the immediate upstream genomic DNA and cDNA sequences flanking the human protein-coding gene TISs and found differential distribution of the human-specific STRs in comparison to the overall distribution of STRs on both platforms. Further, we propose a link between STRs and TIS selection, based on the differential co-occurrence rate of human-specific STRs with non-homologous TISs and non-specific STRs with homologous TISs. The data presented here have implications at the inter-and intraspecies levels, which warrant further functional and evolutionary studies.  Table 1 (a), Table 2 (b), and several randomly selected proteins, whose codons were flanked by non-specific STRs (c), between human and three other species. Maximum identity scores were annotated for each gene in every species. Identity scores were considerably higher for the proteins in the non-specific STR compartment