key: cord-1041969-a5v2ctcv authors: Konishi, T. title: Pattern of COVID-19 epidemics in Japan influenced by the control measures date: 2021-03-26 journal: nan DOI: 10.1101/2021.03.24.21253923 sha: 474d5de2a81b16011dc3b00d14dd1513e9c3c742 doc_id: 1041969 cord_uid: a5v2ctcv Background. COVID-19 has spread worldwide since it appeared in 2019. In contrast to many other countries with epidemics, Japan differed in that it avoided lockdowns and instead asked people for self-restraint. A travel campaign was conducted with a huge budget, but the number of PCR tests is severely limited. These choices may have influenced the course of the epidemic. Methods. The increase or decrease in the classes of SARS-CoV-2 variants was estimated by analyzing the published sequences with an objective multivariate analysis. This methodology does not rely on unverifiable assumptions. It observes the samples in multiple directions, digesting complex differences into simpler forms. The results were compared over time with the number of confirmed cases, PCR tests, and overseas visitors. Results. Three epidemic peaks in Japan have been caused by domestically mutated variants. However, due to the relaxation of immigration restrictions, several infective variants have been imported from abroad and are currently competing for expansion. In other countries, there are few other cases where multiple variants have peaked. The chaotic situation in Japan will continue for a while, in part, because no effort has been made to find asymptomatic carriers and details of the vaccination program are undecided. The worldwide COVID-19 pandemic continues (WHO 2020) , with rapid mutation of variants (Konishi 2021) . In Japan, the epidemic began with spread from a cruise ship. Hitherto, there have been three peaks. The responses to COVID-19 in Japan were different from those of some countries that succeeded in controlling the disease. Strict lockdowns were not carried out; instead, Japan twice issued states of emergency and urged the public to refrain from going out unnecessarily and eating out late at night. These measures were just a 'request for self-restraint', and there were no confirmations, penalties, or compensations. Schools were closed during the first state of emergency but opened during the second. The government launched a travel-promoting campaign "Go To travel", which provides residents with subsidies of up to 50% on transportation, hotels, restaurants, and shopping; 26 billion dollars was the proposed budget. However, the number of public PCR tests is fairly low, and testing is only performed on patients with obvious symptoms. Initially, it was planned to find clusters and infection routes, but this scope was beyond the capacity of the investigation agency and has already been abandoned substantively. Small vendors performing PCR tests are increasing, but the results are not counted in the official records. As of February 2021, COVID-19 has not been contained, people living in major cities are required to voluntarily stay at home, and travel into Japan from abroad is restricted. There are not sufficient facilities to isolate mild and asymptomatic victims. At present, over 500 victims with mild symptoms are waiting at home without medication in Tokyo (Bureau of Social Walfare and Public Health 2021). 110-90% of the beds for severely ill patients in Tokyo hospitals are occupied (NHK 2021) . In this study, we examined this situation with respect to changes in the relative proportions of virus variants. All the sequence data available were observed in an objective manner by considering the sequence matrix as a multivariate variable and applying principal component analysis (PCA) (Jolliffe 2002) . This approach is different than phylogenetic trees (Yang & Rannala 2012) , which require many unverifiable assumptions that reduce the objectivity of analysis (Ellis & Silk 2014) . Since several directions of differences can be observed individually, fine classification can be performed with high reproducibility. These data were observed in chronological order and compared with the number of confirmed cases, the number of overseas visitors, and the number of PCR tests performed. Data source. Nucleotide sequences of 15,746 samples were obtained from GISAID (Elbe & Buckland-Merrett 2017) on February 10, 2021. Sequences were aligned using DECIPHER (Wright 2015), converted to a Boolean vector, and subjected to PCA (Konishi et al. 2019) . Sample principal components (PCs) were scaled based on the length of the sequence (Konishi 2015) . All calculations were performed using R (R Core Team 2020). The ID, acknowledgements, and scaled PC (sPC) of the samples and sPC of bases are available from Figshare (Konishi 2020) . The number of confirmed cases was obtained from the WHO (WHO 2021). The number of PCR tests was obtained from the Ministry of Health, Labour and Welfare (Ministry of Health 2021). The number of foreign visitors was obtained from the official statistics counter (e- Stat 2021) . The number of current cases and PCR tests in the other countries was obtained from the official homepage of the Ministry of Health of the corresponding countries. PCA. This analysis represents the differences among samples of multivariate data through a set of common directions, which are shown as independent vectors (Jolliffe 2002) . Here, the samples are mutations that evolved from an original virus. Hence, the samples will fall into several related groups; each group is different from the others, with a unique direction common to the group. The sequence matrix is converted to a stack of Boolean vectors to allow for calculation (Konishi et al. 2019) . When m is the number of samples (in reality, to cover sequences with n bases, the length of a Boolean vector becomes 5n; here for simplicity, it is described as below), the matrix X is given as follows: Next, the average of the samples, a, is found, and X is centred by subtracting each row with the average: = ( 1 , 2 , ⋯ ), = − . This centred matrix, C, was applied to the PCA. It is subjected to singular value decomposition, = Σ * , where U and V are unitary matrices that specify the directions of the differences. As the character of the unitary matrix, the scale of each column and row of U and V is one, * = , and * = . Their columns can be regarded as vectors that represent the axes of each PC. Σ is a diagonal matrix that records the scaling of each axis in descending order. The PCs for the samples, S, were found as = = Σ. The indicates the rotation of C around the centre a, retaining the shape of C. The results of the rotation are along the axes. This is the same as Σ, which is the unitary matrix given the scale. Each of the columns of S represents the PCs: the leftmost column is PC1; the second column is PC2. The descending character of Σ orders the scaling of PCs. All of the information is conserved, and all the calculation steps are reversible. Once found, axis V can be applied to other sets of matrices (Konishi 2015) . This characteristic is beneficial when applying a classification to newly found samples. PCs for bases, B, can be found in the diagonal direction above, * = Σ * , as = * = Σ. Therefore, S and B are inextricably linked; for example, samples with many positive B in an axis will become highly positive in S along the same axis. On the contrary, the characteristics of a sample that shows a high score on an axis will appear in the same axis as B. To enable comparisons with datasets with different sizes of n or m, PCs can be scaled for different sizes (Konishi 2015) . The scaled versions of S and B, sPC for samples and sPC for bases, are /√ and /√ , respectively. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 For these averages and axes, we used a sequence matrix collected evenly from all over the world as of October 2020 (Konishi 2021) . However, at that point, the new variants of South Africa and England had not yet appeared. Therefore, to find these variants, a new set of axes was used. Both sets of axes can be downloaded from Figshare (Konishi 2020) . The conditions for finding variants from each country are as follows (Konishi 2021) With the first and second axes of PCA, the variants so far were divided into groups of 0 to 3 (Fig. 1A) . These groups were commonly observed in other countries on all continents (Konishi 2021 (Wu et al. 2020; Zhou et al. 2020) , and includes the same variants found on the cruise ship Diamond Princess. Since then, this group has disappeared worldwide, perhaps because of lesser infectivity than other groups (Konishi 2021 ). Subsequently, Group 1 was the mainstream throughout 2020 (Fig. 1B) . This group formed the first peak when it first appeared in Japan. This peak was suppressed by the 1st state of emergency but was not contained. The remaining variant mutated gradually, with changes detected by PC78, and formed the second peak. The second state of emergency was not issued with this peak; hence, many cases remained. Instead, the "Go To travel" campaign was launched. The increasing number of victims enabled other mutations detected by PC28, which caused the early period of the 3rd peak. The travel campaign may have helped spread the variant to a wider . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101/2021.03.24.21253923 doi: medRxiv preprint area of the country, expanding the number of confirmed cases of the 3rd peak. Group 1 remained primary until December 2020, but some infectious variants from overseas had already appeared in that month (Fig. 3A) . In January 2021, when the third wave peaked further, there was a qualitative difference because the number of variants from overseas had increased. Groups 1 and 2 switched positions (Fig. 1B) . The number of people entering Japan was on the rise, with an average of approximately 4 million people per month arriving in Japan in 2019. The first state of emergency reduced this number to 20,000 (Fig. 1B) . However, the number of visitors has gradually increased since then. Visitors were asked to voluntarily quarantine in hotels for two weeks. However, as the number of visitors increased, variants from overseas appeared in Japanese cities. As of January 2021, there were almost the same number of Group 1 and 2 variants (Fig. 1B, 3B ). Group 1 includes the 'new variant' of England (Volz et al. 2021 ). In addition, since December 2020, Group 2 strains from the USA (Konishi 2021) have increased. Similarly, since January 2021, the number of new South African variants has been increasing. In addition to the original strains, Group 0 included those derived from England and the pan-European variant. In addition to Group 1, the original variants in Group 2 gradually changed (Fig. 2) . Both new variants remained even in January 2021 (Fig. 3B) . A large peak of the third wave occurred with a mixture of these multiple variants. Group 1 variants were changing in Japan in 2020. At the end of the year, the number of variants originating overseas has increased, leading to a large third wave. There are still quite a few original Japanese variants, so they may also have high infectivity. Unfortunately, most of the sequences in Japan have no published sampling dates, so a detailed increase or decrease cannot be confirmed or predicted. The differences in government measures between virus-free and epidemic countries is remarkable. To remain clean, it is necessary to lock down until there are no new carriers; once new patients emerge, lockdown to a sufficient scale is required immediately. This approach requires sufficient PCR tests. Otherwise, pathogens that increase exponentially and continue to mutate rapidly cannot be controlled, since an epidemic can occur from a single asymptomatic carrier (Johansson et al. 2021 ). This is a mathematical truth, but it has been backed up by reality. Repeating incomplete lockdown is an error that corresponds to a piecemeal force attack. Japan is a bad example of this approach. Vaccination is another promising strategy, but it has been delayed in Japan, and the program remains undecided (Ministry of Health 2021). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted March 26, 2021. ; Japan is unique in that many variants are simultaneously prevalent. In contrast, one major variant, which is mutated domestically or imported from abroad, is prevalent in many other countries (Konishi 2021) ; for example, the epidemic Group 2 variant in the USA may have mutated domestically and subsequently entered Australia and Japan, causing new epidemics in those countries. The new variant in South Africa may have entered that country from outside. This variant may have mutated in areas without sequencing, since no possible parental variants have been recorded. The earliest record of the pan-European variant was from Spain, which had been occupied by its parental variant. This variant has caused epidemics in many EU countries. The situation in Japan is a result of the fact that the domestic variants were failed to be contained, as well as the importation of highly infectious variants by recklessly loosening the border. The original intention of Japan to block the transmission route and resolve the infection was unsuccessful. Moreover, a partial lockdown that requests only self-suppression failed to contain the virus. As a result of these inappropriate measures, multiple infectious variants have spread and have taken root in Japan. Most people cannot be vaccinated for at least until the next year. If the borders are loosened again, more new variants will emerge, and the rooted variants will be released from the country to the world. Fig. 1 Changes in groups of variants found in Japan. A. The scaled principal component (sPC) of the sequence found in Japan. Axes PC1 and PC2. Four groups are obvious; this framework is common to other countries 8 . Numbers from 0 to 3 are tentatively assigned. Increase/decrease of each group. Group 3 was dominant until March, then Group 1 became dominant, and Group 2 was rapidly increasing in January 2021 (left axis). The green arrows are the periods of the states of emergency, and the black arrow is the Go To travel campaign period, which is currently suspended. The thick grey line is the number of confirmed cases, and the blue line is 1/10 the number of PCR tests (right axis). PCR is performed only on symptomatic patients, which was on average 16 times the number of positives. The grey dotted line is half the number of arriving visitors. It was reduced to 0.5 % of usual numbers but has since increased slightly. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Tokyo Metropolitan Government. 2021. Status of test positives Immigration statistics Data, disease and diplomacy: GISAID's innovative contribution to global health Scientific method: Defend the integrity of physics Principal Component Analysis Principal component analysis for designed experiments SARS-CoV-2 found in each continents Progressing adaptation of SARS-CoV-2 to humans. bioRxiv : the preprint server for biology Principal Component Analysis applied directly to Sequence Matrix About new coronavirus infection New coronavirus: Bed usage rate R: A language and environment for statistical computing Transmission of SARS-CoV-2 Lineage B.1.1.7 in England: Insights from linking epidemiological and genetic data. virologicalorg A new coronavirus associated with human respiratory disease in China Molecular phylogenetics: principles and practice A pneumonia outbreak associated with a new coronavirus of probable bat origin We would like to thank Editage (www.editage.com) for English language editing.