Disease diagnosis of cockatiels
The only gross lesion in the dead cockatiels was weak fibrinous perihepatitis. Escherichia coli was identified in the liver by bacteriological culture. The presence of avian bornavirus, avian polyomavirus, beak and feather disease virus, and psittacid herpesvirus 1 was investigated by PCR, but none of these viruses were detected. Hemagglutination test on harvested allantoic fluid was negative. Microscopic lesions were observed in liver multifocal hepatitis, perivascular lymphocyte and heterophil infiltration, minor pericarditis, and lung congestion.
Metagenomics analysis
Four cockatiel samples (15AD75-1, -2, -3, and -4) were sequenced, yielding 8,160,360 to 14,041,743 reads each. All reads were broadly classified into one of four groups: (1) eukaryotic (95.2–99.9 %), (2) bacterial (<0.1–1.7 %), (3) viral (<0.1 %), and (4) unidentified (<0.1–3.9 %). Homology-based (BLAST) classification of 1,614–144,494 bacterial reads identified four different bacterial species with sequence identity to avian bacteria. The identified bacterial sequences were assigned to known families, including C. jejuni (Campylobacteraceae), C. psittaci (Chlamydiaceae), Escherichia coli (Enterobacteriaceae), and Clostridium colinum. C. jejuni was detected in all four samples, and C. psittaci sequences were detected in three (15AD75-1, -2, and -3). Sequence reads classified as Enterobacteriaceae were identified in sample 15AD75-2, and C. coli sequences were identified in sample 15AD75-1. Although the sample preparation enriched for viruses, viral sequence reads represented fewer than 0.1 percent of reads in three samples and were completely absent in the other sample, 15AD75-4 (Table 1). C. jejuni and C. psittaci were detected in three or four samples. Because these two organisms are pathogens in psittaciformes, we analyzed both in detail.
Comparison of the de novo assembly tools
To obtain more accurate and longer sequence contigs of bacteria, we compared the results of three de novo assemblers (IDBA-UD, MEGAHIT, and metaSPAdes) using reads classified as C. jejuni from sample 15AD75-1 and reads classified as C. psittaci from sample 15AD75-3. The most accurate assembly was generated by metaSPAdes with 0 misassemblies, mismatches per 100 kb (1383.12), and indels per 100 kb (15.4) in C. jejuni. Moreover, metaSPAdes had only one misassembly with mismatches per 100 kb (40.68) and indel per 100 kb (8.62) in C. psittaci. The most inaccurate assembly was generated by MEGAHIT. In C. jejuni, MEGAHIT yielded slightly longer contigs, with N50 of 681, largest alignment of 3,004, and genome fraction of 3.84%, but had the most misassemblies (4). Similarly, in C. psittaci, MEGAHIT yielded slightly longer contigs, with N50 of 1,104, largest alignment of 5,732, and genome fraction of 61.90, but also had most of misassemblies (13). IDBA-UD had intermediate accuracy, with no misassemblies in C. jejuni but 11 in C. psittaci (Table 2).
Next, using sample 15AD75-1, we compared C. jejuni 16S rRNA gene sequences, generated by IDBA-UD, MEGAHIT, and metaSPAdes assembly tools, with the reference sequence of C. jejuni ZJB021 (GenBank accession no. CP048767.1). The sequence aligned by IDBA-UD had a 28 nt insertion at positions 451–478 (GGGAGTAAAGTTAATACCTTTGCTCAT) instead of TTC, as well as various misassemblies, but sequences obtained using Megahit and metaSPAdes had no misassemblies (Figure 2A). We then compared ompA sequences of C. psittaci from sample 15AD75-3, generated by IDBA-UD, MEGAHIT, and metaSPAdes, with the reference sequence C. psittaciGIMC 2005 (GenBank accession no. CP024451.1). Sequences aligned by IDBA-UD and MEGAHIT had many misassemblies at positions 541–994 and consisted of short contigs (615 nt), but the sequence obtained using metaSPAdes had zero misassemblies and was 994 nt in length (Figure 2B). Thus, more accurate and longer sequences were generated by metaSPAdes. These sequences were deposited in the GenBank database under accession numbers MW534394 and MW544064.
Genetic analysis
To genetically characterize the identified bacteria relative to known references, we generated phylogenetic trees of the flaA gene of C. jejuni and the ompA gene of C. psittaci using the neighbor-joining method. Figure 3a shows three clusters of flaA sequences that were not correlated with host or country. C. jejuni15AD75 was closely related to a cluster in genogroup A but was not sub-grouped with other strains. This partial flaA gene (1152 bp) of 15AD75 (accession no. MW544065) had 89.61% nucleotide identity to strain C. jejuni9090 (accession no. CP007181.1), but also had 91.23% nucleotide identity to strain C. coliRM4661 (accession no. CP007181.1). In the recombination analysis using RDP, we detected a significant recombination event between breakpoints (positions 1121 and 1377), with C. coli RM4661 (Turkey, USA) as the minor parent (P-value = 1.351 × 10-8) and C. jejuni 9090 (Human, Slovenia) as the major parent (Figure 4a). Recombination was supported by bootstrap support with a P-value of 3.609 × 10-12 (Figure 4b). 15AD75 had 97.6% nucleotide sequence identity with Campylobacter coli (C. coli) RM4661 (accession no. CP007181.1) and 75.1% identity with C. jejuni 9090 (accession no. CP007181.1) at breakpoints. The breakpoint sequences were closer to C. coli than to C. jejuni.
In the phylogram, ompA of the C. psittaci was clustered in genotype A, the major genotype associated with strains from parrots, chicken, and humans (Figure 3b). C. psittaci15AD75 was similar to reference parrot strains (accession no. MH507065.1, MH507064.1, KR010621.1, MH138297.1, CP003790.1, KR010619.1, KR010620.1), with more than 99.89% sequence identity. In addition, it was similar to human isolates, with 100% and 99.87% identity to strains from Russia (accession no. CP024453.1, CP024451.1, CP024455.1) and Japan (accession no. AB468956.1), respectively.