Data Collection and Sequencing

A total of 1,881 nasal swab and 89 lung tissue samples were collected from swine from 131 pig farms in southern, midwestern, and southeastern Brazil during 2009–2012 (Figure 1). Nursery and growing pigs were selected for nasal swab sampling during 2 cross-sectional studies in 2010 and 2011, where pigs showing typical clinical signs of influenza infection (e.g., fever, labored abdominal breathing, and dyspnea) were preferentially chosen for sampling. Lung samples were collected from pigs of different production phases (suckling, nursery, and fattening) and submitted to a private diagnostic laboratory for necropsy and/or histopathologic examination, and positive lung samples for IAV by immunohistochemical analysis were sent to the virology laboratory at EMBRAPA (Brazilian Agricultural Research Corporation, Concordia, Brazil) for virus isolation, reverse transcription quantitative PCR (RT-qPCR), and sequencing.

IAV was detected from nasal swab samples from pigs on 24 of 62 farms. Fifty-nine (3.13%) of 1,881 nasal swab samples analyzed were positive by RT-qPCR. Moreover, 58 (65.16%) of 89 lung samples with pneumonic lesions, from 44 pig farms, were positive for IAV by RT-qPCR. Viral RNA was extracted from the original samples and tested for the matrix gene of IAV and pH1N1 IAV by 2 separate RT-qPCRs (28,29). Virus was isolated by inoculating lung tissue supernatant or nasal swab samples into MDCK cells or into specific pathogen–free embryonated chicken eggs (30). To confirm virus isolation, we tested the supernatants or allantoic fluids for hemagglutinin (HA) activity and by RT-qPCR. Nasal swab and lung samples were considered negative for IAV if a second passage in specific pathogen–free chicken eggs or in MDCK cells was negative by HA test and RT-PCR. Of the 117 PCR-positive samples, 41 virus isolates were obtained: 14 from nasal swab samples and 27 from lung samples.

We conducted whole-genome sequencing using an ABI 3130xl analyzer (Applied Biosystems, Foster City, CA, USA) and Illumina MiSeq (Illumina, San Diego, CA, USA). PCR was conducted by using a set of primers specific for the amplification of the 8 IAV gene segments (31,32) and a primer set for the amplification of pH1N1 gene segments. For sequencing on an Illumina MiSeq, RT reaction and amplification of the 8 IAV gene segments were conducted by using primers as previously described (32,33). In total, at least partial sequence data were obtained for 35 IAVs. Of these, 16 swIAVs provided whole-genome sequence data of sufficient quality to be included in the final phylogenetic analysis.

Phylogenetic Analysis

Nucleotide alignments were generated for 11 discrete datasets: H1 (human seasonal virus–like, 209 viruses), H1 (pandemic virus–like, 451 viruses), H3 (463 viruses), N1 (pandemic virus–like, 311 viruses), N2 (683 viruses), and the 6 internal gene segments (polymerase basic [PB] 2, PB1, polymerase acidic [PA], nucleoprotein [NP], matrix [MP], and nonstructural [NS]). Each dataset comprised 1) Brazilian swIAVs sequenced for this study; 2) Brazilian swIAVs that we sequenced and published previously (26,27); and 3) related human and swine viruses, collected globally, that were downloaded from the Influenza Virus Resource (34) available in GenBank, which were studied previously (35) (Table 1; Technical Appendix Table). To accommodate relatively long but not fully complete sequences, the H3 alignment was trimmed to 1158 nt and the N2 alignment was trimmed to 1161 nt, and sequences less than this length were excluded.

For each of the 5 sets of sequence data, an alignment was generated by using MUSCLE v3.8.31 (36), with manual correction using the program Se-Al (http://tree.bio.ed.ac.uk/software/seal/). The phylogenetic relationships of each of the 5 datasets were inferred by using the maximum-likelihood (ML) method available in RAxML v7.2.6 (37), incorporating a general time-reversible model of nucleotide substitution with a γ-distributed rate variation among sites. To assess the robustness of each node, we conducted a bootstrap resampling process (500 replicates), again using the ML method available in RAxML v7.2.6. Because of the computational complexity of these processes, we used the high-performance computational capabilities of the Biowulf Linux cluster at the National Institutes of Health (http://biowulf.nih.gov).

Divergence Times

A time-scaled Bayesian approach was used to estimate the timing of the human-to-swine transmission events associated with the H1, H3, and N2 segments, due to the long branch lengths separating the Brazilian swine viruses from the most closely related human seasonal viruses. We used a relaxed uncorrelated lognormal molecular clock, a flexible Bayesian skyline plot (BSP) demographic model (10 piece-wise constant groups), and a general-time reversible model of nucleotide substitution with a γ-distributed rate variation among sites. Markov chain Monte Carlo was run separately 3 times for each dataset for at least 100 million iterations, with subsampling every 10,000 iterations. The analysis used BEAST v1.8.0 (38) and the BEAGLE library (39) to improve computational performance. All parameters reached convergence, as assessed visually by using Tracer v1.6 (http://tree.bio.ed.ac.uk/software/tracer/), with statistical uncertainty reflected by values of the 95% highest posterior density (HPD). The initial 10% of the chain was removed as burn-in, runs were combined by using LogCombiner v1.8.0 (http://beast.bio.ed.ac.uk/logcombiner), and maximum clade credibility (MCC) trees were summarized by using TreeAnnotator v1.8.0 (http://beast.bio.ed.ac.uk/treeannotator).

Identification of New Human-Origin H3N2 IAVs in Swine

On the H3 phylogeny, the 4 H3N2 viruses from swine in Brazil in 2011 are monophyletic, forming a single clade supported by 100% posterior probability (Figure 2) and 100% bootstrap support (Technical Appendix Figure 1): A/swine/Brazil/365-11-7/2011(H3N2), A/swine/Brazil/231-11-1/2011(H3N2), A/swine/Brazil/355-11-6/2011(H3N2), and A/swine/Brazil/365-11-6/2011(H3N2) (Table 1; Technical Appendix Table). This clade is phylogenetically distinct from other swIAVs found in Argentina and North America that also are of human seasonal H3N2 virus origin and appears to be a previously uncharacterized introduction of a human seasonal H3N2 virus into swine (Figure 2; Technical Appendix Figure 1). The human H3N2 viruses that are most closely related to the Brazilian swIAVs were isolated during the late 1990s (e.g., A/Malaysia/13241/1997[H3N2]) and have a long branch length separating the clade of swIAVs from Brazil from the older human virus diversity.

The time-scaled MCC tree (Figure 2) indicates that human-to-swine transmission could have occurred at any time along this long branch between the node representing the estimated time to the most recent common ancestor of the clade of swIAVs from Brazil (2009.6; 95% HPD 2008.5–2010.4) and the node representing the time to the most recent common ancestor for the swine clade and most closely related human virus (1995.5; 95% HPD 1995.1–1995.9). The much higher intensity of sampling of human viruses than swine viruses means that the long branch length is more likely to represent unsampled swine viruses than unsampled human viruses and the actual time of human-to-swine introduction is likely to be closer to 1995.5. Detection of H3N2 swIAVs in 3 Brazilian states (Santa Catarina, Rio Grande do Sul, and Mato Grosso do Sul; Figures 1, 2) is consistent with extensive circulation of this swIAV within swine populations in southern and midwestern Brazil.

Identification of New Human-Origin H1N2 IAVs in Swine

Figure 3

Figure 3. Phylogenetic relationships between human and swine influenza H1 segments. Time-scaled Bayesian maximum clade credibility (MCC) tree inferred for the hemagglutinin (H1) sequences of 209 viruses, including 2 swine viruses from Brazil...

On the H1 phylogeny, the 5 H1N2 viruses collected from swine in Brazil during 2011 also are monophyletic, although the clade is supported by just below the 70% bootstrap threshold on the ML tree (64% bootstrap support, Technical Appendix Figure 2) and 80% posterior probability threshold on the MCC tree (79%, Figure 3). The lower support for this clade appears to be related to the early divergence of these swIAVs into 2 well-supported subclades (100% posterior probabilities). One subclade comprises 3 swIAVs collected in Paraná and Rio Grande do Sul States: A/swine/Brazil/31-11-1/2011(H1N2), A/swine/Brazil/31-11-3/2011(H1N2), and A/wild boar/Brazil/214-11-13D/2011(H1N2) (Table 1; Figures 1, 3). The second subclade comprises 2 swIAVs collected in Santa Catarina State: A/swine/Brazil/185-11-7/2011(H1N2) and A/swine/Brazil/232-11-13/2011(H1N2). The presence of H1N2 viruses from domestic swine and a wild boar within the first subclade is consistent with viral transmission between wild and domestic swine in Brazil.

The estimated time of transmission of the human H1N2 viruses into swine is 2002.1–2003.1 (95% HPD 2001.8–2004.7), which is consistent with the period of global circulation of the unusual reassortant H1N2 virus in humans (2001–2003). The most closely related human virus was isolated in 2002: A/New York/417/2002/H1N2. Because of the low availability of human H1N2 sequences from this period and the low posterior probability (MCC tree) and bootstrap support (ML tree) for the clade of Brazilian H1N2 viruses, we cannot conclude with certainty that the H1N2 swIAVs identified in Brazil represent a single viral introduction from humans into pigs instead of 2 contemporaneous introductions involving similar human H1N2 viruses. The most parsimonious explanation, however, is that a single introduction of a human seasonal H1N2 virus into swine occurred, and shortly thereafter the virus diversified into 2 subclades that co-circulated in swine until at least 2011, when these H1N2 viruses were identified in swine in Brazil.

Two Introductions of Human N2 Segments in Swine

In contrast to the H3 and H1 phylogenies (Figures 2, 3), the Brazilian swIAVs are not monophyletic on the N2 phylogeny, representing 2 different introductions of the N2 segment from human seasonal H3N2 viruses into swine in Brazil (Figure 4). One N2 clade (clade N2-a, Figure 4) is defined by 100% posterior probability support and 90% bootstrap support (Technical Appendix Figure 3) and comprises 6 Brazilian swIAVs: A/swine/Brazil/185-11-7/2011(H1N2), A/swine/Brazil/31-11-1/2011(H1N2), A/swine/Brazil/232-11-13/2011(H1N2), A/swine/Brazil/232-11-14/2011(H1N2), A/swine/Brazil/355-11-6/2011(H3N2), and A/swine/Brazil/365-11-7/2011(H3N2) (Table 1). Within the N2-a clade, the 4 H1N2 viruses cluster separately from the 2 H3N2 viruses as 2 genetically distinct subclades (90% and 100% bootstrap support, respectively; online Technical Appendix Figure 3). The 6 swIAVs in N2-a were collected from 4 Brazilian states: Santa Catarina, Rio Grande do Sul, Paraná, and Mato Grosso do Sul (Figure 1). The other N2 clade (clade N2-b, Figure 4) consists of 2 swIAVs from Brazil of the H1N2 subtype, defined by 100% posterior probability and 90% bootstrap support (Technical Appendix Figure 3): A/wild boar/Brazil/214-11-13D/2011(H1N2) and A/swine/Brazil/31-11-3/2011(H1N2). These swIAVs were collected in Paraná and Rio Grande do Sul States, consistent with dispersal within southern Brazil despite the low number of samples (Figure 1; Table 1).

Both the Brazilian N2-a and N2-b clades are closely related to human seasonal H3N2 viruses that circulated during the late 1990s (e.g., A/New York/521/1998[H3N2] for clade N2-a and A/New York/251/1998[H3N2] for clade N2-b) and therefore have similar estimated times of human-to-swine transmission on the MCC tree: 1997.1–2002.8 (95% HPD 1996.3–2005.6) for clade N2-a and 1997.2–2006.9 (95% HPD 1996.1–2009.0) for N2-b (Figure 4). Again, the higher intensity of influenza surveillance in humans than in swine suggests that the 2 human-to-swine transmission events associated with clades N2-a and N2-b are more likely to have occurred during late 1990s. These estimates overlap with the timing of the introduction of the H3 segment into swine in Brazil (Figure 2), and a parsimonious explanation is that the H3 clade and clade N2-a both represent the same introduction of human H3N2 viruses during the late 1990s, although additional sequencing is needed. In contrast, no evidence exists that the N2 of human H1N2 origin has continued to circulate in swine in Brazil beyond its initial introduction during the early 2000s. Rather, during the decade since the introduction of H1N2 viruses into the swine population, the H1N2 viruses have acquired through reassortment 2 different N2 segments of human H3N2 origin that circulated in swine in Brazil (N2-a and N2-b), representing 2 independent reassortment events between H1N2 and H3N2 viruses.

Introductions of Human Pandemic H1N1 IAVs into Swine in Brazil

The 6 viruses of human pH1N1 origin that we sequenced from swine in Brazil were not monophyletic on the H1 or N1 tree, indicating that these viruses represent multiple separate human-to-swine introductions rather than clonal expansion of a single human pH1N1 introduction into pigs (Table 1; Technical Appendix Figures 4, 5). Additional pH1N1 HA and neuraminidase (NA) sequences were available in GenBank from swine in Brazil; these sequences also were not monophyletic or closely related to the pH1N1 viruses that were sequenced in our study, suggesting additional introductions of human pH1N1 viruses into swine herds in Brazil. The lower phylogenetic resolution of the recently emerged pH1N1 virus and large number of singleton viruses make precise estimation of the number of human-to-swine introductions difficult. As a lower-bound estimate based on monophyletic clusters supported by high bootstrap values (>70) on the pandemic H1, N1, and PB2 trees (Technical Appendix Figures 4–6), we conservatively estimate 8 introductions of pH1N1 from humans into swine in Brazil, including additional sequence data from GenBank (Table 2). Including singleton viruses and poorly supported clusters, as many as 15 putative human-to-swine introductions in Brazil were evident on the HA phylogeny (Technical Appendix Figure 4). Some of the long branch lengths separating singleton swine viruses from human viruses might represent onward transmission of a pH1N1 introduction in the swine population in Brazil, but additional sequencing is needed to confirm sustained circulation. Although the resolution of the phylogeny inferred for the matrix segment is even lower, because of the short sequence length, the swIAVs from Brazil also are not monophyletic on the matrix tree, consistent with multiple introductions of pH1N1 into swine in Brazil (Technical Appendix Figure 7). Phylogenies also suggest that human pH1N1 viruses have been transmitted multiple times to swine in other Latin America countries, including Argentina, Colombia, and Mexico (Table 2; Technical Appendix Figures 4–6). Only in Costa Rica was the pH1N1 population monophyletic with high bootstrap support that suggests a single human-to-swine introduction.

Reassortment with pH1N1 Internal Genes

Several additional introductions of pH1N1 were evident only on the internal gene trees (e.g., PB2; Technical Appendix Figure 6) resulting from reassortment with HA and NA segments of human seasonal virus origin (Tables 1,2). In contrast to the genetic diversity of the HA and NA segments in swine in Brazil, the genetic diversity of the 6 internal gene segments in Brazil was restricted to pH1N1, at least for those for which sequence data were available (Table 1). This pattern indicates that the internal genes of pH1N1 viruses have reassorted with the HA and NA of other swIAVs of human seasonal virus origin circulating in Brazil.