Assessment of databases to determine the validity of β- and γ-carbonic anhydrase sequences from vertebrates.

Background The inaccuracy of DNA sequence data is becoming a serious problem, as the amount of molecular data is multiplying rapidly and expectations are high for big data to revolutionize life sciences and health care. In this study, we investigated the accuracy of DNA sequence data from commonly used databases using carbonic anhydrase (CA) gene sequences as generic targets. CAs are ancient metalloenzymes that are present in all unicellular and multicellular living organisms. Among the eight distinct families of CAs, including α, β, γ, δ, ζ, η, θ, and ι, only α-CAs have been reported in vertebrates. Results By an in silico analysis performed on the NCBI and Ensembl databases, we identified several β- and γ-CA sequences in vertebrates, including Homo sapiens, Mus musculus, Felis catus, Lipotes vexillifer, Pantholops hodgsonii, Hippocampus comes, Hucho hucho, Oncorhynchus tshawytscha, Xenopus tropicalis, and Rhinolophus sinicus. Polymerase chain reaction (PCR) analysis of genomic DNA persistently failed to amplify positive β- or γ-CA gene sequences when Mus musculus and Felis catus DNA samples were used as templates. Further BLAST homology searches of the database-derived “vertebrate” β- and γ-CA sequences revealed that the identified sequences were presumably derived from gut microbiota, environmental microbiomes, or grassland ecosystems. Conclusions Our results highlight the need for more accurate and fast curation systems for DNA databases. The mined data must be carefully reconciled with our best knowledge of sequences to improve the accuracy of DNA data for publication.

There are 12 α-CA isozymes, including CA I-IV, CA VA and VB, CA VI, CA VII, CA IX, and CA XII-XIV, that are expressed in humans [6]. Interestingly, CA XV is the only active CA isozyme known to date that is expressed in several vertebrate species but is lost in human and chimpanzee genomes [7]. In addition to the 13 mammalian α-CA isozymes, there are three acatalytic CA-related proteins (CARPs), including CARP VIII, CARP X, and CARP XI, with crucial physiological roles [8][9][10][11]. α-CAs have been reported from many organisms, including both prokaryotes and eukaryotes [12].
Although β-CAs are present in archaea, bacteria, plants, fungi, protozoans, and insects, there are no reports of β-CAs in any vertebrate species [13,14]. Similarly, γ-CAs are present in many prokaryotes and eukaryotes, such as plants and fungi, whereas they do not exist in any vertebrates according to the current knowledge [15,16]. Incomplete β-CA gene sequences have been identified in the genome of the cephalochordate Branchiostoma floridae (the Florida lancelet), but whether they represent a pseudogene or an incompletely sequenced active gene has not been determined [17]. Some annotated βand γ-CA sequences present in databases have been linked to vertebrate genomes, but in fact, they might have originated from either gut microbiota or other normal flora or even from environmental bacterial contamination. Kraken and Taxoblast are two recently designed ultrafast programs to identify contaminant DNA sequences from metagenomic and genome sequencing databases [18,19]. The main limitation of both methods is the lack of accessibility to a computer or server with enough RAM for quick operation while performing genome blast homology searches.
In this study, we first searched for βand γ-CAs in vertebrates using in silico tools. The results obtained from the NCBI and Ensemble databases led us to perform polymerase chain reaction (PCR) amplifications using mouse and cat genomic DNA as templates. The results indicated that the "vertebrate" βand γ-CA sequences detected from databases were presumably derived from gut microbiota, environmental microbiomes, or grassland ecosystems. This finding emphasizes the importance of fast and accurate biocuration of database sequences.
Our further analysis revealed that the genomic organization of the coding genes for the "vertebrate" βand γ-CA proteins was consistent with the single exonic pattern of coding genes in prokaryotes. In addition, the BLAST homology search analysis decrypted the high percentage of identities (73-100%) between the predicted βand γ-CA protein sequences of vertebrates and some other organisms, which mostly involved prokaryotic species (Table 1).

Molecular analysis of βand γ-CA genes from vertebrates
To investigate whether β-CA or γ-CA genes are truly present in vertebrate genomes, we performed PCR using DNA samples extracted from ear punching specimens of M. musculus and whole blood of F. catus. The first round PCRs with low stringent conditions showed some positive signal for the primer pairs P1 and P3 of F. catus and P5 and P8 of M. musculus (Fig. 4a). Estimation of the PCR product size was conducted based on the product length from Table 2. Because the signal remained weak in most cases, we performed the second round PCR using the PCR amplicons from the first round PCR as templates. The results of the second round of PCR are shown in Fig. 4b. The sequencing results revealed that none of the sequenced PCR products represented the predicted β-CA gene from M. musculus or the γ-CA gene from F. catus.

Discussion
CA genes are widely distributed in species of all life kingdoms. Despite this general concept, βand γ-CA genes have never been reported in vertebrate genomes to the best of our knowledge based on previous literature. Our survey on the βand γ-CA gene sequences of vertebrates presented in public databases in 2017-2020 revealed, however, that some sequences were or are still available, such as β-CA genes from L. vexillifer and M. musculus, as well as γ-CA genes from L. vexillifer. Some data were removed in 2019-2020, such as β-CA genes from P. hodgsonii and H. sapiens, as well as γ-CA genes from P. hodgsonii, X. tropicalis, H. sapiens, F. catus, and R. sinicus. Some new sequences appeared and were annotated on databases in 2019-2020, including β-CA genes from H. comes, H. hucho, and O. tshawytscha, as well as the γ-CA gene from H. comes. At first glance, the reports of "vertebrate" βand γ-CA genes in databases raised our interest as a potentially novel discovery, but  enthusiasm gradually dissipated as most data were discontinued in 2019-2020. The BLAST homology search analysis of the predicted "vertebrate" βand γ-CA protein sequences filtered with the "prokaryota" keyword defined that the discontinued βand γ-CA genes belonged to prokaryotes. The most striking false-positive sequences in databases were originally annotated as human βand γ-CAs, which we defined by the BLAST homology search as Mesorhizobium delmotii enzymes instead of human origin (Table 1). Our results suggest that the predicted "human" βand γ-CAs were derived from bacterial contamination of human DNA samples that caused false interpretation during sequencing. As a sign of improved accuracy, these false-positive data were removed from databases in 2019-2020. Another piece of evidence for the bacterial contamination of DNA samples is the contamination of H. comes sample with Muricauda sp. and Bacteroides sp., both of which are abundantly present in seawater sediments [20,21]. In addition, DNA samples of salmon fishes (H. hucho and O. tshawytscha) can be contaminated with gut microbiota or egg-associated bacterial species, such as Flavobacterium sp., Pseudomonas sp., and Hydrogenophaga sp. [22,23]. Comamonadaceae bacterium from gut microbiota may represent the main source of bacterial contamination for the DNA samples of X. tropicalis [24]. Notably, due to the living habitat of R. sinicus in meadows, scrubs, and grasslands and feeding in these important ecosystems, the contamination of the bat DNA sample was mainly derived from plant species, such as Brassica sp. (cruciferous vegetables), instead of contamination from gut microbiota.
The exon count of the predicted "vertebrate" βand γ-CA genes suggested the presence of only a single exon in each case. This finding also supported the idea that prokaryotes from gut microbiota and environmental microbiome are the major source of contaminants that led to unexpected sequencing results from vertebrate DNA samples [25]. This idea was further supported by our PCR analysis of both mouse and cat genomic DNA samples combined with DNA sequencing, which consistently failed to identify any βor γ-CA sequences in mice and cats.   It is clear that a significant amount of incorrect sequence data on both β-CA and γ-CA genes remain in public databases. Some existing examples are β-CA genes of L. vexillifer, M. musculus, H. comes, H. hucho, and O. tshawytscha and γ-CA genes of L. vexillifer and H. comes. The present findings highlight the importance of database curation efforts to achieve a higher degree of accuracy within a shorter revision time.

Conclusions
Online databases are important sources of information for mining genomic and proteomic data of living organisms. Unfortunately, these databases also include misannotated data to some extent due to microbial or other contamination. We used βand γ-CA gene sequences as bioinformatic tools to demonstrate such contamination in various species. Our findings emphasize the importance of fast and reliable curation for achieving betterquality and more accurate genomic and proteomic data.

Identification of βand γ-CAs
In the first step, the βand γ-CA protein sequences from Escherichia coli (NCBI IDs: WP_000658644.1 and WP_131199889.1, respectively) were used as the query in the Basic local alignment search tool (BLAST) for sequence similarity search analysis through the BLASTP program (https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE= Proteins) of NCBI database [26] and TBLASTN program of Ensembl genome browser 95 (https://asia.ensembl. org/Multi/Tools/Blast?db=core) [27]. We filtered the results using "vertebrata" as the organism name, in which the BLASTP program only searched for βand γ-CA protein sequences within vertebrates. Additionally, we applied the scientific name of defined vertebrates as the filter in the TBLASTN program of Ensembl genome browser 95. The obtained βand γ-CA protein sequences were aligned using the Clustal Omega algorithm (https://www.ebi.ac.uk/Tools/msa/clustalo/) [28].
In the second step, we performed a BLAST homology search analysis on the obtained βand γ-CA protein sequences from vertebrates, in which the results were filtered against "prokaryota" as the organism name. Afterward, exon counts were performed to detect βand γ-CA gene sequences from vertebrates through the gene analysis program of the NCBI database.

Molecular analysis of βand γ-CA genes from vertebrates
We designed eight primer pairs using Primer-BLAST for molecular detection of the β-CA gene from Mus musculus (Mouse) and the γ-CA gene from Felis catus (cat) (four primer pairs for each CA gene) identified through bioinformatic methods (Table 2) [29].
The ear blood samples of one M. musculus and 1 ml EDTA-blood samples of one privately-owned F. catus were collected under the permission of the animal ethical committee of the County Administrative Board of Southern Finland (ESAVI/8321/04.10.07/2017 for the mouse and ESAVI/7482/04.10.07/2015 for the cat) for molecular detection of the predicted β-CA gene of M. musculus and γ-CA gene of F. catus. In the Tampere University's animal facility, mice are routinely earmarked and the same samples were used for genotyping purposes in another project. Written consents were collected from the participating cat owners and samples were collected as a part of the ongoing feline genetic research at Dr. Lohi's laboratory. Cats visited a veterinary clinic for a routine sample collection. Genomic DNA was extracted from white blood cells using a semiautomated Chemagen extraction robot (PerkinElmer Chemagen Technologie GmbH, Baeswieler, Germany) according to the manufacturer's instructions. The DNA concentrations were measured using a Qubit fluorometer (Thermo Fisher Scientific, Waltham, Massachusetts, USA) and a Nanodrop ND-1000 UV/Vis Spectrophotometer (Nanodrop Technologies, Wilmington, Delaware, USA), and samples were stored at − 20°C. Polymerase chain reaction (PCR) was performed according to the protocol used by Zolfaghari Emameh R et al. [30]. PCR amplification was run on a thermocycler (Bioer XP Cycler, Hangzhou Bioer Technology Co. Ltd., Hangzhou, China) according to the following details: 95°C (3 min), [95°C (15 s), 60°C (15 s), 72°C (15 s)] × 40 cycles, 72°C (2 min). The amplified products were run on a 1.6% agarose gel and purified using a NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel). The second round of PCR was run as previously described, and the selected PCR amplicons ( Fig. 4; samples 3, 4, 8, and 9) were treated with Exo I and Fast AP enzymes and sequenced using ABI PRISM BigDye® Terminator v3.1 Cycle Sequencing kit and 3500xL Genetic Analyzer (Applied Biosystems, Inc., Foster City, CA, U.S.A.).