Main

An outbreak of serious pneumonia disease was reported in Wuhan, China, on 30 December 2019. The causative agent was soon identified as a novel coronavirus1, which was later named SARS-CoV-2. Case numbers grew rapidly from 27 in December 2019 to 3,090,445 globally as of 30 April 20203, leading to the declaration of a public health emergency, and later a pandemic, by the WHO (World Health Organization). Many of the early cases were linked to the Huanan seafood market in Wuhan city, Hubei province, from where the probable zoonotic source is speculated to originate2. Currently, only environmental samples taken from the market have been reported to be positive for SARS-CoV-2 by the Chinese Center for Disease Control and Prevention4. However, as similar wet markets were implicated in the SARS outbreak of 2002–20035, it seems likely that wild animals were also involved in the emergence of SARS-CoV-2. Indeed, a number of mammalian species were available for purchase in the Huanan seafood market before the outbreak4. Unfortunately, because the market was cleared soon after the outbreak began, determining the source virus in the animal population from the market is challenging. Although a coronavirus that is closely related to SARS-CoV-2, which was sampled from a Rhinolophus affinis bat in Yunnan in 2013, has now been identified6, similar viruses have not yet been detected in other wildlife species. Here we identified SARS-CoV-2-related viruses in pangolins smuggled into southern China.

We investigated the virome composition of pangolins (mammalian order Pholidota). These animals are of growing importance and interest because they are one of the most illegally trafficked mammal species: they are used as a food source and their scales are used in traditional Chinese medicine. A number of pangolin species are now regarded as critically endangered on the International Union for Conservation of Nature Red List of Threatened Species. We received frozen tissue samples (lungs, intestine and blood) collected from 18 Malayan pangolins (Manis javanica) during August 2017–January 2018. These pangolins were obtained during anti-smuggling operations performed by Guangxi Customs officers. Notably, high-throughput sequencing of the RNA of these samples revealed the presence of coronaviruses in 6 out of 43 samples (2 lung samples, 2 intestinal samples, 1 lung–intestine mixed sample and 1 blood sample from 5 individual pangolins; Extended Data Table 1). With the sequence read data, and by filling gaps with amplicon sequencing, we were able to obtain six complete or near complete genome sequences—denoted GX/P1E, GX/P2V, GX/P3B, GX/P4L, GX/P5E and GX/P5L—that fall into the SARS-CoV-2 lineage (within the genus Betacoronavirus of the Coronaviridae) in a phylogenetic analysis (Fig. 1b). The genome sequence of the virus isolate (GX/P2V) has a very high similarity (99.83–99.92%) to the five sequences that were obtained through the metagenomic sequencing of the raw samples, and all samples have similar genomic organizations to SARS-CoV-2, with eleven predicted open-reading frames (ORFs) (Fig. 1a and Extended Data Table 2; two ORFs overlap). We were also able to successfully isolate the virus using the Vero E6 cell line (Extended Data Fig. 1). On the basis of these genome sequences, we designed primers for quantitative PCR (qPCR) detection to confirm that the raw samples were positive for coronavirus. We conducted further qPCR testing on another batch of archived pangolin samples collected between May and July 2018. Among the 19 samples (9 intestine tissues, 10 lung tissues) tested from 12 animals, 3 lung tissue samples from 3 individual pangolins were positive for coronavirus.

Fig. 1: Evolutionary relationships among sequences of human SARS-CoV-2, pangolin coronaviruses and the other reference coronaviruses.
figure 1

a, Genome organization of coronaviruses including the pangolin coronaviruses obtained in this study, with the predicted ORFs shown in different colours (ORF1a is omitted for clarity). The pangolin coronavirus strain GX/P2V is shown with its sequence length. For comparison, the human sequences NC_045512.2 and NC_004718.3, and bat sequences MG772933.1, GQ153541.1 and KC881006.1 are included (see Extended Data Table 6 for sources). b, Phylogeny of the subgenus Sarbecovirus (genus Betacoronavirus; n = 53) estimated from the concatenated ORF1ab, S, E, M and N genes. Red circles indicate the pangolin coronavirus sequences generated in this study (Extended Data Table 1). GD/P1L is the consensus sequence re-assembled from previously published raw data7. Phylogenies were estimated using a maximum likelihood approach that used the GTRGAMMA nucleotide substitution model and 1,000 bootstrap replicates. Scientific names of the bat hosts are indicated at the end of the sequence names, and abbreviated as follows: C. plicataChaerephon plicata; R. affinis, Rhinolophus affinis; R. blasii, Rhinolophus blasii; R. ferrumequinum, Rhinolophus ferrumequinum; R. monoceros, Rhinolophus monoceros; R. macrotis, Rhinolophus macrotis; R. pearsoni, Rhinolophus pearsoni; R. pusillus, Rhinolophus pusillus; R. sinicus, Rhinolophus sinicus. Palm civet (P. larvata, Paguma larvata; species unspecified for Civet007 and PC4-13 sequences) and human (H. sapiens, Homo sapiens) sequences are also shown.

In addition to the animals from Guangxi, after the start of the SARS-CoV-2 outbreak researchers of the Guangzhou Customs Technology Center re-examined five archived pangolin samples (two skin swabs, two unknown tissue samples and one scale) obtained in anti-smuggling operations performed in March 2019. Following high-throughput sequencing, the scale sample was found to contain coronavirus reads, and from these data we assembled a partial genome sequence of 21,505 bp (denoted as GD/P2S), representing approximately 72% of the SARS-CoV-2 genome. Notably, this virus sequence, obtained from a pangolin scale sample, may in fact be derived from contaminants of other infected tissues. Another study of diseased pangolins in Guangdong performed in 2019 also identified viral contigs from lung samples that were similarly related to SARS-CoV-27. Different assembly methods and manual curation were performed to generate a partial genome sequence that comprised 86.3% of the full-length virus genome (denoted as GD/P1L in the phylogeny shown in Fig. 1b).

These pangolin coronavirus genomes have 85.5% to 92.4% sequence similarity to SARS-CoV-2, and represent two sub-lineages of SARS-CoV-2-related viruses in the phylogenetic tree, one of which (comprising GD/P1L and GD/P2S) is very closely related to SARS-CoV-2 (Fig. 1b). It has previously been noted that members of the subgenus Sarbecovirus have experienced widespread recombination8. In support of this, a recombination analysis (Fig. 2) revealed that bat coronaviruses ZC45 and ZXC21 are probably recombinants, containing genome fragments derived from multiple SARS-CoV-related lineages (genome regions 2, 5 and 7) as well as SARS-CoV-2-related lineages, including segments from pangolin coronaviruses (regions 1, 3, 4, 6 and 8).

Fig. 2: Recombination analysis.
figure 2

a, Sliding window analysis of changing patterns of sequence similarity between human SARS-CoV-2, pangolin and bat coronaviruses. The potential recombination breakpoints are shown in pink dash lines, and regions separated by the breakpoints are alternatively shaded in yellow. These potential breakpoints subdivide the genomes into eight regions (regions with fewer than 200 bp were omitted), indicated by the red bars at the bottom of the analysis boxes. The names of the query sequences are shown vertically to the right of the analysis boxes. The similarities to different reference sequences are indicated by different colours. Guangdong pangolin-CoV GD/P1L and pangolin-CoV GD/P2S were merged for this analysis. The blue arrows at the top indicate the position of the ORFs in the alignment. b, Phylogenetic trees of different genomic regions. SARS-CoV- and SARS-CoV-2-related lineages are shown in blue and red tree branches, respectively. Branch supports obtained from 1,000 bootstrap replicates are shown. Branch scale bars are shown as 0.1 substitutions per site.

More notable, however, was the observation of putative recombination signals between the pangolin coronaviruses, bat coronavirus RaTG13 and human SARS-CoV-2 (Fig. 2). In particular, SARS-CoV-2 exhibits very high sequence similarity to the Guangdong pangolin coronaviruses in the receptor-binding domain (RBD) (97.4% amino acid similarity, indicated by red arrow in Fig. 2a; the alignment is shown in Fig. 3a), even though it is most closely related to bat coronavirus RaTG13 in the remainder of the viral genome. Indeed, the Guangdong pangolin coronaviruses and SARS-CoV-2 possess identical amino acids at the five critical residues of the RBD, whereas RaTG13 only shares one amino acid with SARS-CoV-2 (residue 442, according to numbering of the human SARS-CoV9) and these latter two viruses have only 89.2% amino acid similarity in the RBD. Notably, a phylogenetic analysis of synonymous sites only from the RBD revealed that the topological position of the Guangdong pangolin is consistent with that of the remainder of the viral genome, rather than being the closest relative of SARS-CoV-2 (Fig. 3b). Therefore, it is possible that the amino acid similarity between the RBD of the Guangdong pangolin coronaviruses and SARS-CoV-2 is due to selectively mediated convergent evolution rather than recombination, although it is difficult to differentiate between these scenarios on the basis of the current data. This observation is consistent with the fact that the sequence similarity of ACE2 is higher between humans and pangolins (84.8%) than between humans and bats (80.8–81.4% for Rhinolophus sp.) (Extended Data Table 3). The occurrence of recombination and/or convergent evolution further highlights the role that intermediate animal hosts have in the emergence of viruses that can infect humans. However, all of the pangolin coronaviruses identified to date lack the insertion of a polybasic (furin-like) S1/S2 cleavage site in the spike protein that distinguishes human SARS-CoV-2 from related betacoronaviruses (including RaTG13)10 and that may have helped to facilitate the emergence and rapid spread of SARS-CoV-2 through human populations.

Fig. 3: Analysis of the RBD sequence.
figure 3

a, Sequence alignment showing the RBD in human, pangolin and bat coronaviruses. The five critical residues for binding between SARS-CoV RBD and human ACE2 protein are indicated in red boxes, and ACE2-contacting residues are indicated by yellow boxes as previously described9. In the Guangdong pangolin-CoV sequence, the codon positions encoding the amino acids Pro337, Asn420, Pro499 and Asn519 have ambiguous nucleotide compositions, resulting in possible alternative amino acids at these sites (threonine, glycine, threonine and lysine, respectively). Sequence gaps are indicated with dashes. The short black lines at the top indicate the positions of every 10 residues. GD, Guangdong; GX, Guangxi. b, Phylogenetic trees of the SARS-CoV-2-related lineage estimated from the entire RBD region (top) and synonymous sites only (bottom). Branch supports obtained from 1,000 bootstrap replicates are shown. Branch scale bars are shown as 0.1 substitutions per site.

To our knowledge, pangolins are the only mammals in addition to bats that have been documented to be infected by a SARS-CoV-2-related coronavirus. It is notable that two related lineages of coronaviruses are found in pangolins that were independently sampled in different Chinese provinces and that both are also related to SARS-CoV-2. This suggests that these animals may be important hosts for these viruses, which is surprising as pangolins are solitary animals that have relatively small population sizes, reflecting their endangered status11. Indeed, on the basis of the current data it cannot be excluded that pangolins acquired their SARS-CoV-2-related viruses independently from bats or another animal host. Therefore, their role in the emergence of human SARS-CoV-2 remains to be confirmed. In this context, it is noteworthy that both lineages of pangolin coronaviruses were obtained from trafficked Malayan pangolins, which originated from Southeast Asia, and that there is a marked lack of knowledge of the viral diversity maintained by this species in regions in which it is indigenous. Furthermore, the extent of virus transmission in pangolin populations should be investigated further. However, the repeated occurrence of infections with SARS-CoV-2-related coronaviruses in Guangxi and Guangdong provinces suggests that this animal may have an important role in the community ecology of coronaviruses.

Coronaviruses, including those related to SARS-CoV-2, are present in many wild mammals in Asia5,6,7,12. Although the epidemiology, pathogenicity, interspecies infectivity and transmissibility of coronaviruses in pangolins remains to be studied, the data presented here strongly suggests that handling these animals requires considerable caution and their sale in wet markets should be strictly prohibited. Further surveillance of pangolins in their natural environment in China and Southeast Asia are necessary to understand their role in the emergence of coronaviruses and the risk of future zoonotic transmissions.

Methods

Data reporting

No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.

Ethics statement

The animals studied here were rescued and treated by the Guangxi Zhuang Autonomous Region Terrestrial Wildlife Medical-aid and Monitoring Epidemic Diseases Research Center under the ethics approval (wild animal treatment regulation No. [2011] 85). The samples were collected following the procedure guideline (Pangolins Rescue Procedure, November, 2016).

Sample collection, viral detection and sequencing of pangolins in Guangxi

We received frozen tissue samples of 18 pangolins (M. javanica) from Guangxi Medical University, China, that were collected between August 2017 – January 2018. These pangolins were seized by the Guangxi Customs during their routine anti-smuggling operations. All animal individuals comprised samples from multiple organs including lungs, intestine and blood, with the exception of six individuals for which only lung tissues were available, five with mixed intestine and lung tissues only, one with intestine tissues only, and one comprising two blood samples. Using the intestine–lung mixed sample we were able to isolate a novel Betacoronavirus using the Vero-E6 cell line (from ATCC; Extended Data Fig. 1). The cell line was subjected to species identification and authentication by microscopic morphologic evaluation and growth curve analysis, and was tested free of mycoplasma contamination. The cell line was not on the list of common misidentified cell lines by ICLAC. A High Pure Viral RNA Kit (Roche) was used for RNA extraction on all 43 samples. For RNA sequencing (GX/P2V and GX/P3B), a sequencing library was constructed using an Ion Total RNA-Seq Kit v2 (Thermo Fisher Scientific), and the library was subsequently sequenced using an Ion Torrent S5 sequencer (Thermo Fisher Scientific). For other samples, reverse transcription was performed using an SuperScript III First-Strand Synthesis System for RT–PCR (Thermo Fisher Scientific). DNA libraries were constructed using the NEBNext Ultra II DNA Library Prep Kit and sequenced on a MiSeq sequencer. The NGS (next-generation sequencing) QC Toolkit V2.3.3 was used to remove low-quality and short reads. Both BLASTn and BLASTx were used to search against a local virus database, using the data available at NCBI/GenBank. Genome sequences were assembled using the CLC Genomic Workbench v.9.0. To fill gaps in high throughput sequencing and obtain the whole viral genome sequence, amplicon primers based on the bat SARS-like coronavirus ZC45 (GenBank accession number MG772933) sequence and the coronavirus contigs obtained in the initial sequencing were designed for further amplicon-based sequencing.

A total of six samples (including the virus isolate) contained reads that matched members of the genus Betacoronavirus (Extended Data Table 1). We obtained near complete viral genomes from these samples (98%, compared to SARS-CoV-2), which were designated GX/P1E, GX/P2V, GX/P3B, GX/P4L, GX/P5E and GX/P5L. Their average sequencing coverage ranged from approximately 8.4X to 8,478X (Extended Data Fig. 2a-f). On the basis of these genome sequences, we designed primers for qPCR to confirm the positivity of the original tissue samples (Extended Data Table 4). This revealed an original lung tissue sample that was also qPCR positive, in addition to the six original samples with coronavirus reads. We further tested an additional 19 samples (nine intestine tissues and ten lung tissues), from 12 smuggled pangolins sampled between May–July 2018 by the group from Guangxi Medical University. The genome sequences of GX/P1E, GX/P2V, GX/P3B, GX/P4L, GX/P5E and GX/P5L have been submitted to GISAID database and assigned accession numbers EPI_ISL_410538 - EPI_ISL_410543.

Sample collection, viral detection and sequencing of pangolins in Guangdong

After the start of the SARS-CoV-2 outbreak, the Guangzhou Customs Technology Center re-examined their five archived pangolin samples (two skin swabs, two unknown tissue and one scale) obtained in anti-smuggling operations undertaken in March 2019. RNA was extracted from all five samples (Qiagen), and was subjected to high-throughput RNA sequencing on the Illumina HiSeq platform by Vision medicals. The scale sample was found to contain coronavirus reads using a BLAST-based approach. These reads were quality assessed, cleaned and assembled into contigs by both de novo (MEGAHIT v1.1.313) and using reference (BWA v0.7.1314) assembly methods, using BetaCoV/Wuhan/WIV04/2019 as a reference. The contigs were combined, and approximately 72% of the coronavirus genome (21,505 bp) was obtained. This sequence has about 6.6× sequencing coverage (Extended Data Fig. 2g) and denoted pangolin-CoV GD/P2S. This sequence has been deposited on GISAID with accession number EPI_ISL_410544.

A recently published meta-transcriptomic study of pangolins7 deposited 21 RNA-seq raw files on the SRA database (https://www.ncbi.nlm.nih.gov/sra). We screened these raw read files using BLAST methods and found that five (SRR10168374, SRR10168376, SRR10168377, SRR10168378 and SRR10168392) contained reads that mapped to SARS-CoV-2. These reads were subjected to quality assessment, cleaning and then de novo assembly using MEGAHIT13 and reference assembly using BWA14. These reads were then merged and curated in a pileup alignment file to obtain the consensus sequences. This combined consensus sequence is 25,753 bp in length (about 86.3% of BetaCoV/Wuhan/WIV04/2019; about 6.9× coverage) and denoted pangolin-CoV GD/P1L (available in the Supplementary Information Dataset). Notably, it has 66.8% overlap and a sequence identity of 99.79% with the GD/P2S sequence. As the genetic distance between these viruses is very low, for the recombination analysis we merged the GD/P1L and GD/P2S sequences into a single consensus sequence to minimize gap regions within any sequences.

The viral genome organizations of the Guangxi and Guangdong pangolin coronaviruses were similar to SARS-CoV-2. They possessed nine non-overlapping open reading frames (ORFs) plus two overlapping ORFs, and shared the same gene order of ORF1ab replicase, envelope glycoprotein spike (S), envelope (E), membrane (M), nucleocapsid (N), plus other predicted ORFs. A detailed comparison of the ORF length and similarity with SARS-CoV-2 and bat coronavirus RaTG13 is provided in Extended Data Table 2.

Sequence, phylogenetic and recombination analyses

The human SARS-CoV-2 and bat RaTG13 coronavirus genome sequences were downloaded from Virological.org (http://virological.org) and the GISAID (https://www.gisaid.org) databases in January 2020, with the data kindly shared by the submitters (Extended Data Table 5). Other coronaviruses (subgenus Sarbecovirus) were downloaded from GenBank (Extended Data Table 6) and compared to those obtained here. We constructed a multiple sequence alignment of their complete genomes and individual genes using MAFFT v.7.27315. Maximum likelihood phylogenies were estimated using RAxML v.8.2.1216 from 100 inferences, using the GTRGAMMA model of nucleotide substitution with 1,000 bootstrap replicates. To investigate potential recombination events, we used SimPlot v.3.5.117 to conduct a window sliding analysis to determine the changing patterns of sequence similarity and phylogenetic clustering between the query and the reference sequences. A full plot for the recombination analysis is provided in Extended Data Fig. 3. We also examined phylogenetic clusters performed directly from the multiple sequence alignment. Maximum likelihood trees were estimated from each window extraction (that is, genome regions 1 to 8) using RAxML as described above.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.