Our investigation included 50 COVID-19 patients; 54.0% were women, and the mean age was 55.2 (range 22–96) years (Table 1). Of the case-patients, we categorized 3 cases as imported because the patients stayed in Wuhan before traveling to Hong Kong in mid-January. Four patients traveled to Japan and other provinces of China in mid-January and were hospitalized after they returned to Hong Kong, but active community transmission of COVID-19 was not officially reported in these areas during the study period, so these cases were considered possible local infections. The other 43 patients’ cases were categorized as local infection because of no recent travel history.

Eighteen (36.0%) patients had chronic illnesses, of which cardiovascular and cerebrovascular diseases were the most common (Table 1). In total, 74.0% of the patients were experiencing cough at admission. Fever occurred in 58.0% of patients at time of admission, but that rate gradually increased to 64.0% during the course of hospitalization. Other less common symptoms were muscle aches (25.0%), sore throat (24.0%), shortness of breath (24.0%), and diarrhea (14.3%) (Table 2). Two persons (4.0%) were asymptomatic throughout the study period. On radiologic examination, 27 (54.0%) had bilateral pneumonia, 11 (22.0%) had unilateral pneumonia, and 17 (34.7%) showed multiple areas of mottling and ground-glass opacity. None of the patients were co-infected with other respiratory viruses or fungi.

Intensive care unit admission was relatively uncommon (3/50, 6.0%) in our cohort compared with admission rates in previous studies (19–22). This difference could be attributed to underdiagnosis of milder cases during the initial COVID-19 outbreak in China. One patient’s sputum specimen was culture-positive for Klebsiella aerogenes bacteria; a second patient’s specimen was culture-positive for Ralstonia pickettii bacteria. Both had acute respiratory distress syndrome and acute respiratory injury accompanied by septic shock or acute renal injury and required admission to the intensive care unit.

Figure 1

Figure 1. Demographics of coronavirus disease patients, Hong Kong, February 2020. Rectangular boxes indicate men or boys and stadium-shaped boxes indicate women or girls. Travel history within the 14-day incubation period before...

Of the 50 case-patients, 42 (84.0%) could be clustered based on their epidemiologic links (Figure 1). We identified 6 transmission clusters (clusters 1–6). Cluster 1 involved a 4-member family. The father, who traveled to Guangdong, China, in late January 2020, was believed to have infected his wife and subsequently their daughter and son-in-law at a family gathering. Clusters 2 and 3 were family clusters of local infection and unknown source. Both clusters involved 3 household members with no recent travel history. Cluster 4 was attributed to a superspreading event (SSE): a barbecue and hotpot party involving 19 family members in late January. Symptom onset in these patients occurred during days 2–13 after the party. A colleague of 1 infected person, who did not attend the party, also tested positive for SARS-CoV-2. Cluster 5 initiated from a resident of a public housing estate, in whom COVID-19 was diagnosed on January 30. Eleven days later, diagnoses were made in 3 members of a household that resided in the same building (10 stories below) the index case-patient. Two household members subsequently attended a family gathering of 29 persons at a Chinese restaurant during the incubation period. COVID-19 was diagnosed consecutively in 3 persons ≈2 weeks after the gathering. In addition, a Filipino domestic aide of 1 infected family member, who did not attend the family gathering, also tested positive. For cluster 6, the first reported case was in a 70-year-old woman who visited a Buddhist worship hall during the Chinese New Year. A further 8 persons who visited the same Buddhist worship hall during this period later tested positive for SARS-CoV-2. At the data cutoff point, >4 other household members who had never been to the worship hall also tested positive. Details of the demographic and epidemiologic information on the cases and clusters are provided in Appendix 1 Tables 2–8.

Consensus genomes of all 50 cases were constructed based on nanopore sequencing and refined by Illumina sequencing. On average, 62,387 reads/genome were obtained with 550× coverage for nanopore, and 18,747 reads/genome were obtained with 132× coverage for Illumina platform. The consensus genome size was ≈29.9 kbp with GC content ≈38%. The genomes were highly conserved with the first SARS-CoV-2 genome and had an average sequence identity of 99.98% (range 99.94%–100.0%). We identified 64 nonsynonymous substitutions from all 50 genomes (Appendix 1 Table 9). Orf3a-G251V was the most frequent amino acid substitution; 44/50 (88.0%) of the samples harbored this mutation, after which Orf1ab-H3233Y (30/50, 60.0%) and S-L8V (27/50, 54.0%) were most common.

Genomewide single-nucleotide polymorphisms were used to contextualize phylogenetic placement of Hong Kong strains in SARS-CoV-2 global phylogeny (Appendix 2 Figure 1). However, because the samples were taken at the early stage of global outbreak, the genetic variability between strains was limited, resulting in several unresolved branches and marginal supporting bootstrap values. Nevertheless, when compared with SARS-CoV-2 strains isolated from other regions, Hong Kong strains tended to aggregate mainly in 2 lineages. Lineage 1 consisted of 4 Hong Kong strains that clustered with most isolates from China (n = 32). Lineage 2, which consisted of 44 Hong Kong strains, was more closely related to strains seen in South Korea (n = 7) and France (n = 5).

Figure 2

Figure 2. Maximum-likelihood phylogenetic tree of 50 coronavirus disease cases, Hong Kong, February 2020. The tree was rooted on the earliest published genome of severe acute respiratory syndrome coronavirus 2 (GenBank accession...

In examining the phylogeny of the COVID-19 outbreak in Hong Kong, we identified 2 distinctive groups (Figure 2). The first group consisted of 2 imported cases and the cases in cluster 1. The second group originated with a single robust node with bootstrap value of 94% and a common mutation Orf3a-G251V. The second group could be further separated into 3 subgroups. The first subgroup mainly consisted of the cases in cluster 5, the public housing estate–related SSE. The second subgroup included cases in cluster 4 associated with the family hotpot party, cluster 3, and 2 isolated cases (case 23 and case 43). These samples shared the same missense mutations at S-L8V and Orf1ab-H3233Y. Finally, the third subgroup included the cases from cluster 6, an SSE originating from a Buddhist worship hall, in which Orf1ab-G295V were identified.

According to Bayesian time-scaled phylodynamic analysis, strict clock and relaxed clock models estimated the time of most recent common ancestor of COVID-19 outbreak in Hong Kong as December 24, 2019 (95% Bayesian credible interval [BCI] December 11, 2019–January 5, 2020). The evolutionary rate was 3.04 × 10⁻³ substitutions/site/year (95% BCI 2.04–4.09 × 10⁻³ substitutions/site/year) (Appendix 2 Figure 2). Based on demographic data, the average time from symptom onset to hospital admission was ≈8.5 days. The estimated reproduction number was calculated at 1.84 (95% BCI 1.37–2.35).

This study provides a territorywide overview of early COVID-19 outbreak in Hong Kong, an international city with borders connecting to mainland China, by integrating demographic, clinical, epidemiologic, phylogenomic, and phylodynamic data. In Hong Kong, most cases recorded in January 2020 were imported cases. After February 1, most were local cases and close contacts of those case-patients, indicating local community transmissions. Transmission in closed settings, especially during family and religious gatherings, is a hallmark of recent cases recorded in Hong Kong. Among 6 clusters identified on the basis of epidemiologic links, 3 (clusters 4–6; Figure 1) were considered SSEs because of the larger number of persons involved (n = 8–13). We performed whole-genome sequencing on all 50 cases to investigate phylogenetic relationship and transmission link.

The SARS-CoV-2 samples in Hong Kong had 99.98% identity to the reference genome (GenBank accession no. NC_045512.2) and demonstrated no apparent major genome modification since the initial COVID-19 outbreak in Wuhan. As shown in global phylogeny, SARS-CoV-2 genomes isolated in Hong Kong could be segregated into 2 lineages. Lineage 1 was phylogenetically related to the strains isolated from China and was the cause of the cases in Cluster 1. Lineage 2 was more closely related to strains from France and South Korea. It also harbored a common mutation at Orf3a-G251V, which accounted for 88.0% of cases in this study.

Regarding the local phylogenetic analysis, clustering of samples was highly concordant to the epidemiologic link, despite the marginally supportive bootstrap value of the nodes because of the limited genetic variability. Cluster 1 demonstrated the closest genetic distance to the reference genome among all cases reported in Hong Kong (Figures 1, 2). The index case of cluster 1 (case 66) was initially defined as possible local infection because the patient traveled to Guangdong Province, which was not considered to have active community transmission at that time. However, our sequencing result demonstrated that the genome of case 66 was 100% identical to the first published SARS-CoV-2 genome, and all cases in cluster 1 did not harbor Orf3a-G251V, which was recognized as a hallmark of local cases with unknown source in our community. Therefore, instead of possible local infections, cluster 1 was more likely imported from mainland china through index case-patient 66.

Cluster 5 originated within a public housing estate, in which a family of 3 members (cases 42, 48, and 49) were suspected to have been infected through a confirmed case-patient (case 12) who lived 10 stories above them in the same building, through a potentially faulty sewage pipe setup or other environmental exposure. Based on phylogenetic analysis, viral genomes in cluster 5 shared a similar genetic distance from the reference genome and were assigned to the same branch of the tree. This finding supports a potential transmission link among these cases.

Figure 3

Figure 3. Geographic distribution of 50 coronavirus disease cases, Hong Kong, February 2020.Geographic information is marked according to the residence of the index case-patient in each cluster. Clusters known to...

Cluster 4 was a family gathering–associated SSE during Chinese New Year. In concert with epidemiologic information, all 11 cases from cluster 4 shared 3 common missense mutations, namely S-L8V, Orf1ab-H3233Y, and Orf3a-G251V; 7 cases shared identical genomes. Considering the fast-evolving property of RNA viruses, the presence of identical genetic sequences among the strains implies that transmission occurred over a short period or even in a single event. Meanwhile, 2 isolated cases (case 23 and case 43) and 3 cases from another local cluster (case 38, case 39, and case 40) shared highly similar genomes to those of cluster 4 (Figures 1, 2). Although no apparent epidemiologic links were observed, the high degree of genomic similarity suggests that these cases might have originated in a single source. That speculation was further supported by the geographic distribution of case-patients who lived near one another and whose social circles might have overlapped (Figure 3). Our results demonstrate that the integration of epidemiologic and phylogenetic data is critical for providing more accurate information about transmission patterns.

Cluster 6 was an SSE occurring in a Buddhist worship hall. Two missense mutations, Orf1ab-G295V and Orf1ab-L3606F, were unique to this cluster. Epidemiologic investigation identified a 43-year-old monk (case 102; Figure 1), who was the abbot of the worship hall and had traveled to mainland China in early January. He was sent to a quarantine center in late February after being linked to a series of confirmed cases connected to the worship hall. He was asymptomatic throughout the study period. Phylogenetic analysis showed that this case was closest to the root of the cluster (Figure 2), suggesting that case 102 could be the index patient of cluster 6. By the time of data cutoff, the cluster involved 13 patients and spread was ongoing. This pattern demonstrates the possibility of a hidden spreader as a source of COVID-19 community outbreak. That likelihood also highlights the importance of rapid quarantine of close contacts of confirmed case-patients, regardless of the presence of symptoms, to halt community spread.

In the evolutionary clock study, the reproduction number of COVID-19 within Hong Kong as of February 28, 2020, was estimated at 1.84 (95% BCI 1.37–2.35). That value strongly indicated that the outbreak in Hong Kong was ongoing, but it was smaller than the estimated reproduction number of 2.6 in Wuhan (23,24). The smaller value is a combined outcome of reduced growth rate and increased δ. The reduced growth rate is attributed to strong public health awareness among the general public, which resulted in greatly reduced social activities and strong compliance with mask-wearing during this period (25,26). The increased δ can be attributed to robust laboratory surveillance and fast quarantine time. In addition, time of most recent common ancestor for the cases in Hong Kong was determined to be December 24, 2019, ≈25 days before the first patient in our cohort (Case 2) demonstrated symptoms on January 18, 2020 (Figure 1).

Our study has several limitations. Although we included 53.8% of the cases reported in Hong Kong as of February 28, another 43 cases, including 2 fatal cases, were not analyzed in this study. Moreover, incubation periods of cases in which the source of infection is unknown might vary widely. Studies have demonstrated that incubation periods can vary from 4.5 to 15.8 days (24) and can be longer for patients experiencing mild symptoms. However, because patients might already be infectious during the incubation period, the reproductive number in this study could be underestimated. Furthermore, our calculations were based solely on phylodynamic analysis, which could differ from calculations on the basis of epidemiologic models. Finally, ambiguous bases were observed in some of our consensus genomes. This ambiguity is mainly because whole-genome sequencing was performed on respiratory specimens instead of viral culture, in which viral load plays a critical role in the subsequent genome quality as reflected by the cycle threshold of each specimen. The paucity of viral load in specimens could affect the yield of sequencing libraries. In our study, specimens with cycle threshold <28 were usually free of ambiguous bases. Nevertheless, the uncovered area only accounted for ≈1–3% of the entire viral genome, although the remaining mapped regions had an average coverage of >100×, which should provide sufficient and accurate information for subsequent analyses (Appendix 1 Table 10).

In conclusion, phylogenomic data were consistent with epidemiologic findings that transmission in closed settings, especially during family and religious gatherings, is a hallmark of COVID-19 outbreak in Hong Kong. Social distancing and vigilant infection control measures, such as rapid isolation of suspected or confirmed case-patients and their close contacts, are crucial for containing COVID-19 in the community.