Introduction

Every gene has its unique and distinct genomic location. However, two adjacent genes are often observed to share their coding sequences with each other, i.e., their coding sequences overlap partially or entirely. For instance, in E. coli enterotoxin gene astA was shown to be completely embedded in transposase-like gene IS1414 (McVeigh et al. 2000). Several genomic analyses have inferred that these overlapping genes are present in a wide range of taxonomic groups including viruses (Chirico et al. 2010; Pavesi 2006; Pavesi et al. 2013; Rancurel et al. 2009; Simon-Loriere et al. 2013) prokaryotes (Cock and Whitworth 2007, 2010; Johnson and Chisholm 2004; Sabath et al. 2008; Sakharkar et al. 2005), plants (Kunova et al. 2012; Quesada et al. 1999) animals, (Makalowska et al. 2007; Sanna et al. 2008; Veeramachaneni et al. 2004) and fungi (Gerads and Ernst 1998; Williams et al. 2005). Moreover, the existence of overlapping genes has been experimentally validated in prokaryotes as well as in eukaryotes (McVeigh et al. 2000; Szklarczyk et al. 2007).

Overlapping genes can be classified broadly into two types: (1) same strand overlapping genes, which are transcribed from the same strand of DNA (also known as parallel or unidirectional overlaps); (2) different strand overlapping genes that are transcribed from two opposite strands of DNA (also known as anti-parallel overlaps) (Sanna et al. 2008). Anti-parallel overlaps are further subdivided into convergent (−><−) and divergent types (<− −>). Convergent type involves the overlapping of 3′ ends and is also denoted as tail-to-tail overlaps, whereas divergent type entails the overlapping of 5′ ends and is often denoted as head-to-head overlap. Different species have different abundance and distributions of parallel and anti-parallel overlaps (Sanna et al. 2008). In bacterial genome, same strand overlaps are more common (Johnson and Chisholm 2004), whereas genomes of higher eukaryotes such as human and mouse configure 90 % of overlapping genes in anti-parallel orientation (Sanna et al. 2008).

There could be varied reasons behind the fixation and maintenance of gene overlaps in different species. The potential role of OGs in gene expression regulation has been demonstrated in Arabidopsis genome (Kunova et al. 2012). An overlapping gene astA that entirely overlapped with IS1414, encode heat-stable enterotoxin 1 (EAST 1) in enteroaggregative E. coli (EAEC). This further emphasizes the role of overlapping genes in bacterial pathogenesis as well (McVeigh et al. 2000). It has already been mentioned that OGs in prokaryotic genomes may cause genome compaction (Sakharkar et al. 2005). However, in viruses, other than genome compaction, OGs may create a new gene that encodes a novel protein necessary for viral adaptation to host by a mechanism, commonly known as “overprinting” (Pavesi 2006). In addition, translational coupling of functionally related genes are favored by gene overlaps in the phage genome (Inokuchi et al. 2000).

Prokaryotes are often subjected to environmental changes. Changes in habitat temperature is one such environmental challenges that prokaryotes are frequently exposed to. High temperature often denatures protein (Lepock et al. 1988), disrupts membrane structure and integrity (Baatout et al. 2005) and hampers cell homeostasis. Therefore, high temperatures could pose severe threat to cell survival. In order to combat this adverse or stressful condition, some prokaryotes have evolved several molecular strategies. These include: increase in purine content of mRNA (Lao and Forsdyke 2000; Das et al. 2006; Lambros et al. 2003), rRNA secondary structure stability (Galtier and Lobry 1997), reduction of protein structural disorder (Burra et al. 2010) and more importantly, the reduction of genome size that facilitates thermophilic adaptation of bacteria (Sabath et al. 2013). Bacterial overlapping genes originate through simple extension or elongation of one coding frame into another due to the absence of stop codon in the coding region (Fukuda et al. 1999). Recently, it was proposed that prokaryotes could reduce their genome size by elimination of intergenic regions (genomic streamlining) in response to rise in growth temperature (Sabath et al. 2013). It has already been mentioned that OGs facilitate genome compaction (Sakharkar et al. 2005). However, no studies have ever linked OG content of prokaryotic genome to their growth temperatures. Hence, we intended to investigate whether there is an abundance of OGs in thermophilic prokaryotes compared to non-thermophilic ones. An in-depth comparison of OGs between 50 thermophilic and 206 non-thermophilic genomes revealed that OGs have significant contribution in thermophilic stress tolerance of prokaryotic genomes which leads to their enrichment in thermophilic genomes. Moreover, we have also illustrated the different mechanisms of long (7–50 nucleotides) and short (1–4 nucleotides) overlaps in acclimatizing higher temperature. Thus our study would surely help in better understanding of the mechanisms of thermophilic stress tolerance.

Materials and methods

Retrieval of genomic information

The dataset of this study comprises 256 unique prokaryotes for which, gene overlap data as well as Optimal Growth Temperature (OGT) data were available (Additional file 1). OGT data were obtained from NCBI database (ftp://www.ncbi.nlm.nih.gov/genomes/genomeprj_archive/) and supplementary dataset of Wang et al. (2006). Overlapping gene data were retrieved from Pairwise neighbours database (Palleja et al. 2009). Overlap frequency was calculated as the total number of adjacent gene pairs that overlap divided by total number of adjacent gene pairs in that genome (Sabath et al. 2008). Long overlaps were defined as overlaps spanning a region of 7–50 nucleotides, whereas short overlaps are defined as overlaps spanning a region of 1–4 nucleotides (Fonseca et al. 2014). Spacer length data were extracted from pairwise neighbours database. Proportion of the intergenic region (IG %) was calculated as sum of all spacer length between all adjacent gene pairs divided by genome size.

Genome size information for each genome was obtained from (ftp://www.ncbi.nlm.nih.gov/genomes/genomeprj_archive/lproks_0.txt).

Average Codon Adaptation Index (CAIavg) values of each prokaryotic genome were taken from the supplementary material of Botzman and Margalit (2011).

Calculation of overlap formation frequency

For the calculation of overlap formation frequency in response to change in optimal growth temperature, we considered nine thermophilic–mesophilic pairs that have been previously studied by McDonald (2010). Members of these nine thermophilic–mesophilic pairs were phylogenetically close, exhibited similar GC content but differ in their habitat temperature (Table 2). Additionally, we chose one mesophilic–psychrophilic pair to study the changes in overlap frequency during cold stress. One-to-one orthologous genes between the members of those meso-thermo and meso-psychro pairs were retrieved from OMA genome browser (Altenhoff et al. 2011). For each genome we first estimated the number of OGs that have detectable orthologs in the paired genome. Next, we calculated the number of newly formed overlapping gene in a genome as the number of genes that rendered overlapping in that genome but whose orthologs rendered non-overlapping in the paired genome. Overlap formation frequency of each genome in each pair was estimated as the ratio of number of newly formed OGs divided by the number of OGs that have detectable orthologs in its counter genome. Overlap formation frequency in thermophilic, mesophilic and psychrophilic were denoted as OGthermo, OGmeso, OGpsychro, respectively.

Statistical analyses

All statistical analyses were conducted using the software SPSS version 13. All types of overlap frequencies were found to be non-parametric in Shapiro–Wilk test (P < 0.05). Following non-parametric distribution we used Mann–Whitney U test to detect significant difference in the distribution of different variables between two groups. Spearman’s correlation tests were performed to analyze the correlations between different variables.

Results

Thermophiles have higher OG frequency compared to non-thermophiles

Here, we found a strong negative correlation between OG frequency and the percentage of intergenic DNA in a genome (IG%) (Spearman’s ρ = −0.328; P = 10−6; N = 256). Hence, it is logical to hypothesize that the presence of OGs could be a critical feature of thermophilic prokaryotes. To validate our hypothesis, we conducted a study on 256 distinct prokaryotes (Additional file 1) out of which 50 were thermophiles (optimal growth temperature ≥45 °C) and 206 non-thermophiles (temperature ranges from 16.5 to 42 °C). We calculated overlap frequency for each genome mentioned above and compared the mean overlap frequency of thermophiles with that of non-thermophiles. The results of Mann–Whitney U test including statistical parameters, P value and mean overlap frequency of both the groups (thermophiles and non-thermophiles) were enlisted in Table 1. Our results show that thermophiles have an elevated frequency of gene overlaps than non-thermophiles. We also observed a weak but significant correlation between optimal growth temperature and gene overlap frequency (Spearman’s ρ = 0.197; P = 1.3 × 10−3; N = 256). This observation prompted us to perform further investigation on the role of OGs in acclimatizing thermal stress.

Table 1 Detailed results of Mann–Whitney U test for comparison of the overlap frequency between thermophilic and non-thermophilic groups
Full size table

In order to study how shift in growth temperature modulates overlap frequency in prokaryotic genomes, we estimated gene overlap formation frequencies in nine thermophilic–mesophilic genome pairs that have been previously studied by McDonald (2010) (details given in “Materials and methods” section). Additionally, we chose one meso-psychro pair to study the changes in gene overlap formation frequency during cold stress. Our detailed analysis revealed that in eight out of nine meso-thermo pairs overlap formation frequency in thermophilic genomes (OGthermo) was significantly higher than overlap formation frequency in mesophilic genomes (OGmeso) (Table 2). For the meso-psychro pair, OGmeso was also found to be significantly higher than OGpsychro (overlap formation frequency in psychrophilic genomes) (Table 2).

Table 2 Overlap formation frequency in nine thermophilic–mesophilic pairs and one mesophilic–psychrophilic pair
Full size table

Effect of long and short overlaps in regulating thermophilic stress

We compared the distribution of short (1–4 nucleotides) and long overlap (7–50 nucleotides) frequency between non-thermophiles and thermophiles in our dataset. We obtained a very weak, but significant difference of short overlap frequency between non-thermophiles and thermophiles, whereas the difference of long overlap frequency was quite pronounced in thermophiles compared to non-thermophiles (Table 3). Hence, it would be interesting to examine whether short and long overlaps differ in their impact on the process of acclimatization to a higher temperature range. Long overlap frequency (7–50 nucleotides) was found to yield a significant and robust correlation with optimal growth temperature (Spearman’s ρ = 0.489; P = 10−6; N = 256) (Fig. 1) while short overlap frequency (1–4 nucleotides) did not hold any significant correlation with optimal growth temperature (Spearman’s ρ = 0.062; P = 0.263; N = 256) (Fig. 2). In order to test how LOF is associated with the degree of thermophilicity, we considered their correlation in thermophilic and non-thermophilic groups separately. LOF was found to correlate with OGT in both thermophilic and non-thermophilic group (Spearman’s ρ Thermo = 0.704; P = 10−6; N = 50 and Spearman’s ρ Non-thermo = 0.320; P = 3 × 10−6; N = 206) (Fig S1 and S2, additional file2). Here, we also noticed that LOF and OGT shares significant correlation in archae (37 genomes) (Spearman’s ρ archae = 0.754; P = 10−6) and eubacterial domains (219 genomes) (Spearman’s ρ eubacteria = 0.404; P = 10−6). In our dataset, many overlaps were detected to exceed 50 nucleotides in length. So, we were also curious to study the relationship of these types of overlaps (>50 nucleotides) with temperature. We have also found a significant correlation between very long overlap frequency (>50 nucleotides) and OGT (Spearman’s ρ = 0.276; P = 8 × 10−5; N = 256). Interestingly, comparison of the correlation coefficients between long (7–50 nucleotides) as well as very long overlaps (>50 nucleotides) with OGT, using Steiger’s Z test has shown that long overlap frequency has a more profound effect over temperature compared to very long overlaps (>50 nucleotides) (Z = 3.92; P < 0.01).

Table 3 Summary of Mann–Whitney U test for difference of LOF and SOF between thermophiles and non-thermophiles
Full size table
Fig. 1
figure 1

Scattered plot showing the correlation between LOF and OGT. Here, we plotted the frequency of long overlaps (LOF) with the optimal growth temperature (OGT) of 256 prokaryotic genomes

Full size image
Fig. 2
figure 2

Scattered plot showing the correlation between SOF and OGT. Here, we plotted the frequency of short overlaps (SOF) with the optimal growth temperature (OGT) of 256 prokaryotic genomes

Full size image

Next, we analyzed the role of long and short overlap frequencies in the process acclimatization to high temperature. Since, genome size compaction is an exclusive phenomenon of thermophilic stress tolerance, we decided to focus on the impact of long and short overlap frequency over genome size. We have performed a non-parametric Spearman’s correlation test between genome size and long overlap frequency (LOF) as well as short overlap frequency (SOF). We found a significant negative correlation between LOF and genome size (Spearman’s ρ = −0.548; P = 10−6; N = 256) which indicates that LOF has a significant contribution in genome compaction. Surprisingly, we obtained a positive correlation between SOF and genome size (Spearman’s ρ = 0.168; P = 0.007; N = 256). It was also proposed that genome size in prokaryotes associates positively with genomic GC content (Hildebrand et al. 2010). Furthermore, the short overlap frequency in prokaryotic genomes increases gradually with genomic GC content (Fonseca et al. 2014). Hence, we found it essential to investigate the correlation between genome size and both long and short overlap frequency after controlling for genomic GC content. In both the cases, i.e., LOF and SOF yielded a negative correlation with genome size after controlling GC content (Spearman’s ρ LOF = −0.424; P = 10−5; r 2 = 0.180 Spearman’s ρ SOF = −0.277; P = 1 × 10−4; r 2 = 0.077; N = 256). It is apparent from the r 2 values of the two correlations that LOF accounts for 18 % variation of genome size and SOF accounts for 7.7 % variation of genome size. Therefore, our results indicate that both short and long overlaps participate in genome compaction but the effect of long overlaps on genome size were far more pronounced compared to short overlaps.

OG frequency and other salient factors influencing optimum growth temperature

According to Sabath et al. (2008) the variability of codon usage may influence OG frequency in prokaryotic genomes (Sabath et al. 2008). Consequently, Botzman and Margalit (2011) pointed out that growth temperature has a robust impact over codon usage bias of prokaryotic genomes (Botzman and Margalit 2011). In the previous section, OG frequency was found to be correlated with both intergenic regions (IG%) and growth temperature (OGT). Hence, it would be interesting to explore whether CAIavg and IG% has any influence over both long and short overlaps, respectively. Table 4 clearly shows that both SOF and LOF hold significant correlation with CAIavg and IG%. Abundance of short overlap frequency in thermophilic genomes even in the absence of any direct correlation with growth temperature suggests that the short overlap frequency may have guided by any other confounding factors. Therefore, high CAIavg and low IG% may instigate the rise SOF in thermophilic genomes. To quantitate and evaluate the contribution of each factor related to growth temperature in our study, we did a multivariate linear regression analysis. Table 5 delineates the result of this multivariate regression analysis. The result indicates that the genomic factors studied here guide growth temperature in the following order: long overlap frequency (LOF) > CAIavg > genome size. We found no significant effect of IG% and SOF on optimal growth temperature. So, from our study it is evident that the long overlap frequency is the strongest potential factor regulating OGT.

Table 4 Summary of Spearman’s correlations (correlation coefficient and P value) of LOF and SOF with (a) temperature (b) codon adaptation index (CAIavg) (c) proportion of intergenic DNA (IG %)
Full size table
Table 5 Summary of multiple linear regression with optimum growth temperature as dependent variable and other factors as independent variable
Full size table

Discussion

This study reveals a robust association between overlapping gene content and optimal growth temperature of prokaryotic genomes. The probability of gene overlapping is markedly increased when two neighboring genes are brought closer to each other with minimal intergenic region. This is the reason why overlapping genes are very common in operons where genes are arranged co-directionally, separated by short intergenic spacers (Moreno-Hagelsieb and Collado-Vides 2002). Previously, it has been observed that thermophiles have more structured genomic architecture, where genes are very frequently organized into operons (Yoon et al. 2011). These operons are seldom disrupted by genetic rearrangements in thermophiles (Yoon et al. 2011). These observations prompted us to study whether there is any association between OG content and OGT in prokaryotes. We found a significant difference of overlap frequency between thermophiles and non-thermophiles. In addition, a significant strong correlation between overlap frequency and OGT was also observed. Moreover, multivariate regression analysis revealed a robust impact of LOF over OGT. This further implied that like many other genomic features, OG content is determined by the environmental factor like OGT and increase in OG content could be an effective strategy of prokaryotes in combating thermal stress.

Many theoretical studies as well as experimental evidences suggest that the last universal common ancestor (LUCA) of both archae and bacteria lived in (hyper) thermophilic environment (Akanuma et al. 2013; Brooks et al. 2004; Di Giulio 2003). Therefore, we find it interesting to investigate whether during the evolution of thermophilic ancestor to mesophilic or psychrophilic successor; there exist a selection against the trait of gene overlapping. In order to address this issue, we have chosen ten prokaryotic pair, members of nine of these pairs were one mesophilic and one thermophilic and the tenth pair had one mesophilic and one psychrophilic member. We took orthologous genes between members of each pairs and conducted an in-depth analysis on the conservation of overlapping relationship between the members of each pair. Here, it was noticed that in the course of evolution of prokaryotes from high growth temperature to relatively colder one, frequency of overlapping gene is markedly reduced. Moreover, our results revealed that members of mesophilic and psychrophilic genomes have consistently lower overlap formation frequency than their thermophilic counterpart. Thus, these results imply that in comparison to the genomes of higher growth temperature, evolutionary pressure for overlap formation is generally relieved in the genomes of low optimal growth temperature.

Length of gene overlap shows a wide variation in prokaryotes (Fonseca et al. 2014). Earlier, Fonseca et al. 2014 reported that, in prokaryotes, selection acts against long overlaps while short overlaps are profusely present in the prokaryotic genomes. Another interesting observation in our study with respect to overlap length is the association of long overlap frequency with OGT which is stronger and more robust than total overlap frequency. It was reported that mechanism of thermophilic adaptation differs between archae and eubacteria (Mizuguchi et al. 2007). Hence, we also investigated whether association between OGT and LOF varies between archae and eubacteria. It was found that association between LOF and OGT were significant in both domains of archae and eubacteria. This further shows that although different mechanisms of thermal stress tolerance exist between archae and eubacteria, an elevated overlap frequency was common between the two superkingdoms. Moreover, our results revealed that frequency of long overlaps consistently increases with increase in optimal growth temperature. Thus, it suggests that long overlap frequency changes with the degree in thermophilicity in prokaryotic genomes. But, strikingly short overlap frequency yielded no such direct correlation with OGT. Therefore, we wondered for an explanation of such an observation. Studies on viral system revealed that the overlapping region encoding simultaneously for two protein products are under stronger selective constraints (Simon-Loriere et al. 2013). Moreover, it has been shown that genes that overlap through their entire length (internal overlaps) are evolutionarily more conserved than the genes that overlap partially (terminal overlaps) (Simon-Loriere et al. 2013). Thus, the length of gene overlap could be regarded as an important factor to modulate the selective constraints on overlapping genes. It is evident that mutations that are neutral or nearly neutral under optimal physiological conditions could become deleterious at high temperature and they are commonly called temperature-sensitive mutations (Drake 2009). For this reason, number of studies have reported that coding regions of thermophilic genomes undergo lower rate of base substitution than the coding regions of non-thermophilic genomes (Drake 2009; Friedman et al. 2004). Therefore, it is logical to hypothesize that the increased selective pressure on the overlapping region may favor overlapping genes to be more abundant in thermophilic genomes compared to non-thermophilic genomes. Moreover, due to more stringent selection on long overlaps, these types of overlaps may have been favored in thermophilic genomes as compared to short overlaps. However, further studies are required to assess the effect of overlap length on overall base substitution rate of a given gene in prokaryotic genomes. In our study short overlaps are found to hold a strong correlation with CAIavg. Since, CAIavg has a broad role in thermophilic adaptation (Botzman and Margalit 2011), it might be possible that along with long overlaps, short overlaps helps in survival at higher temperature through increasing CAIavg. Long overlaps (7–50 nucleotides) has a more pronounced effect over genome size compaction than short overlaps and for this reason they may also be more abundant in thermophiles compared to short overlaps and share a robust correlation with OGT.

From our study, it is evident that genome compaction is the primary reason of association of overlapping gene content to thermophily. However, genome size reduction does not necessarily involve an increase in OG frequency. Genomes of endosymbionts often are of small size but contain large intergenic regions due to their functional constraints (Degnan et al. 2011), and hence, contain limited repertoire of gene overlap. Previous studies (Kelkar and Ochman 2013; Sakharkar and Chow 2005; Sakharkar et al. 2004) showed that loss of genes could result in shortening of genome length. Thermophiles, in contrast increases frequency of gene overlapping in their genomes and shorten intergenic region to reduce their genome sizes. In connection to this, here, we would like to draw an important example of cell size reduction in response to high temperature in marine planktonic bacteria (Chrzanowski et al. 1988), where gradual shrinkage of cell volume was found to be an intrinsic property of the cells in response to rise in OGT. More importantly, genome streamlining is also observed in these groups of marine planktonic bacteria (Swan et al. 2013). Hence, further studies are necessary to explore whether OGT has any impact over cell size reduction in overall prokaryotic world, and if it is then the genome compaction (through rise in OG) could be a primary prerequisite to accommodate the genome into a smaller cellular space.

In summary, our observations and interpretations shed light into a relatively unrecognized facet of genomic adaptation of prokaryotes to extreme temperature, where we find an essential and nontrivial connection between overlapping gene content and thermophily. Our study will surely pave inroads to future research on prokaryotic adaptation to extreme temperature.