Trevor Bedford Profile picture
Scientist @fredhutch, studying viruses, evolution and immunity. Collection of #COVID19 threads here: https://t.co/Yc4fun5rcp
Michael Davlin Profile picture eDo Profile picture We are nowhere Profile picture Jim Hanna Profile picture Bagwon Pro American Profile picture 183 added to My Authors
11 May
The drivers of the #COVID19 epidemic in India are certainly multifactorial, but we have now seen the viral lineage B.1.617 linked to this epidemic continue to increase in frequency in India and spread rapidly outside of the country. 1/10
Looking within India there are three primary viral lineages of consequence: B.1.1.7 (in blue) and B.1.351 (in green) introduced into India repeatedly from outside the country and B.1.617 (in yellow) emerging endogenously from within India (nextstrain.org/ncov/asia?c=em…). 2/10
Tracking frequencies over time in sequence data shared to @gisaid shows a continued increase in B.1.617, while recent weeks have shown a decline in B.1.1.7. 3/10
Read 10 tweets
7 May
From Aug 2020 to Mar 2021, the lagged case fatality rate (CFR) of the US #COVID19 epidemic had remained largely constant at ~1.5% and provided a simple method to predict subsequent deaths from current cases. 1/6
I've rerun the previous analysis correlating state-level reported cases with state-level reported deaths with different lags. Using @CDCgov data since Aug 2020, I find that a 19 day lag of cases to deaths maximizes average state-level correlation coefficient. 2/6
This shows the resulting projection for deaths where the gray dashed line shows a lookahead projection where 1.5% of reported cases result in reported deaths 19 days later. This can be compared to the solid red line showing realized 7-day average of reported deaths. 3/6
Read 6 tweets
3 May
Just as we can decompose the US #COVID19 epidemic into a B.1.1.7 epidemic and a non-B.1.1.7 epidemic, we can further partition by variants of concern B.1.1.7, B.1.351 and P.1, where it's clear that P.1 has been gaining ground. 1/13
Here, using data from @GISAID, we see that in terms of frequencies across the US, P.1 has been undergoing more rapid logistic growth in frequency than B.1.1.7, while B.1.351 has been slower than B.1.1.7. 2/13
I'm plotting this with the unusual "logit" y-axis (with 1%, 10%, 50%, etc...) because a straight line in logit space is indicative of logistic growth. This sort of plot makes it easy to compare logistic growth rate of frequency between lineages with different frequencies. 3/13
Read 13 tweets
26 Apr
There are effectively two #COVID19 epidemics in the US at this moment; one largely resolving epidemic comprised of non-variant viruses and one growing epidemic of B.1.1.7. Together they have resulted in a near-plateau of cases throughout much of the spring. 1/10
If we look at virus frequencies in the US using data in @GISAID, we can see that the 7-day weighted frequency of B.1.1.7 has been growing consistently since January and is now over 50% in the US. 2/10
This pattern is repeated across individual states. These six were chosen as states with plentiful genomic data and to provide geographic diversity. B.1.1.7 is dominating throughout the US, except for New York and surroundings where B.1.526 is co-circulating. 3/10
Read 10 tweets
23 Apr
It's hard for me to infer the degree to which new variants are driving the surge in cases in India, but we are seeing rapid growth in frequency of multiple viral variants. 1/5
Here is a @nextstrain view of @GISAID data that focuses on viruses from India and highlights emerging lineages B.1.1.7 (in blue), B.1.351 (in green) and B.1.617 (in orange). Interactive version at nextstrain.org/ncov/asia?c=em…. 2/5
We can fit a logistic growth model to the full genomic dataset from India for these three lineages, where we see logistic growth as "linear on a logit scale". Each of these lineages is estimated to have similar logistic growth rates of ~0.3 per week. 3/5
Read 5 tweets
23 Apr
When variants of concern were first identified in late Dec, the US was not where it needed to be in terms of genomic surveillance. However, with considerable ramp up by the CDC, state labs and academic groups, we now have a remarkable genomic surveillance system. 1/14
My favorite metric for genomic surveillance is the number of cases that have been sampled, sequenced and shared publicly to @GISAID in the previous 30 days. By incorporating both sequencing volume and turnaround time, it tells you how much is known about current circulation. 2/14
Throughout the fall, the US had just 100-300 genomes available that were sampled, sequenced and shared in the previous 30 days. 3/14
Read 14 tweets
14 Apr
Since their recognition in the UK, South Africa and Brazil in Dec 2020 and Jan 2021, the variant of concern lineages B.1.1.7, B.1.351 and P.1 have continued to spread throughout the world with B.1.1.7 so far the most successful of the three. 1/15
These lineages first received attention due to large numbers of mutations to the spike protein along with rapid increases in frequency in the UK, South Africa and Manaus, Brazil, but much subsequent attention has focused on key mutations E484K and N501Y. 2/15
This figure shows genotype at sites 484 and 501 mapped onto a reference phylogeny of ~4k viruses sampled from all over the world with 484K viruses in light orange, 501Y viruses in blue (including B.1.1.7) and 484K+501Y viruses in dark orange (including B.1.351 and P.1). 3/15
Read 16 tweets
18 Feb
After a ~2 month plateau from mid-Nov to mid-Jan, the US #COVID19 epidemic has undergone a steady week after week decline and is now back to daily case counts last seen in late October. A thread on what we might expect going forwards. 1/13
Working with case counts from @COVID19Tracking and Rt estimates from epiforecasts.io, I'm showing US confirmed cases broken out by state alongside transmission rate as measured by Rt through time. 2/13
Generally, Rt > 1 in Nov and Dec corresponding to rising cases and drops below 1 in Jan corresponding to falling cases. We've seen a steady decline in Rt from Nov to Feb. Thus, current decline is not a sudden shift in circumstance, but resulted from reaching Rt < 1. 3/13
Read 13 tweets
3 Feb
With emerging variants of SARS-CoV-2 and initial evidence of antigenic evolution, I've seen comparisons here to seasonal influenza and its rate of evolution. In this thread, I want to ground these comparisons with some data. 1/18
If we follow a transmission chain of SARS-CoV-2 from person to person, we'll generally see one mutation occur across the viral genome roughly every two weeks. 2/18
Here I use data from @nextstrain and @GISAID to compare sampling date to the number of mutations across the SARS-CoV-2 genome relative to initial genomes from Wuhan. This shows a steady accumulation of mutations through time with the average virus now bearing ~24 mutations. 3/18 Image
Read 19 tweets
20 Jan
Important new study by Wibmer et al (biorxiv.org/content/10.110…) of neutralization by convalescent sera on wildtype vs 501Y.V2 variant viruses circulating in South Africa. It shows that mutations present in 501Y.V2 result in reduced neutralization capacity. 1/10
Here, I've replotted data from the preprint to make effect size a bit more clear. Each line is sera from one individual tested against wildtype virus on the left and 501Y.V2 variant virus on the right. Note the log y axis (as is common with this type of data). 2/10
It's clear that 501Y.V2 often results in reductions of neutralization titer, quantified as "fold-reduction" where, for example, a 2-fold reduction in titer would mean that you need twice as much sera to neutralize the same amount of virus in the assay. 3/10
Read 10 tweets
18 Jan
At this point, the countries with most genomic data to analyze spread of the variant virus belonging to cov-lineages.org B.1.1.7 lineage or @nextstrain clade 20I/501Y.V1 are the UK, Denmark and the USA. Here I compare growth rates of B.1.17 across these countries. 1/13
Working from @GISAID data, the UK has 18776 genomes, Denmark has 6089 genomes and the USA has 3093 genomes from specimens collected after Dec 15, 2020. Here, I'm looking at daily genomes with collection dates up to Jan 6 that were not pre-screened by "S dropout". 2/13
For the UK, we can see a steady increase in the frequency of sequenced variant viruses belonging to the B.1.1.7 lineage, reaching ~70% frequency at the end of December. Solid line is a 7 day sliding window average. 3/13
Read 14 tweets
14 Jan
After ~10 months of relative quiescence we've started to see some striking evolution of SARS-CoV-2 with a repeated evolutionary pattern in the SARS-CoV-2 variants of concern emerging from the UK, South Africa and Brazil. 1/19
In SARS-CoV-2, the viral spike protein and in particular the receptor binding domain (RBD) is a locus for important viral evolution and is the primary target for the human immune response (figure from science.sciencemag.org/content/367/64…). 2/19 Image
There had been little evolution in the RBD until ~Oct 2020 when we saw RBD mutations start to spread. Perhaps chief among mutations of interest are E484K and N501Y which mutate nearby sites in the RBD. The evolution of these sites can be seen here: nextstrain.org/ncov/global?c=…. 3/19 Image
Read 19 tweets
29 Dec 20
With data that has emerged in the last week, I'm now 80-90% convinced that infections by the UK variant virus (Pangolin lineage B.1.1.7, @nextstrain clade 20B/501Y.V1) result in, on average, more onward infections, ie are more transmissible. 1/10
My thinking primarily comes from three data points:
1. rapid increase in frequency of variant over wildtype
2. higher secondary attack rate of variant than wildtype
3. increased viral loads of variant over wildtype as measured by Ct
2/10
For point 1 (increase in frequency) we have pretty much the same data as of a week ago, where we see increasing frequency of variant over wildtype across the UK. This can be readily seen in this analysis by @TWenseleers. 3/10
Read 11 tweets
28 Dec 20
Given that the US has not detected SARS-CoV-2 variant viruses 501Y.V1 or 501Y.V2, what bounds can we place on their current frequency in the US based on sample counts? 1/12
So far, the US has not detected any cases that genetically match either the UK variant virus 501Y.V1 (nextstrain.org/ncov/europe?c=…) or the South African variant virus 501Y.V2 (nextstrain.org/ncov/africa?c=…). 2/12
However, because the US is generally slower at turnaround of specimens into SARS-CoV-2 sequences than the UK, we lack confidence that 501Y.V variants are absent from the US. 3/12
Read 13 tweets
23 Dec 20
Given the large discrepancy in specimens collected in Dec that were sequenced and shared between the US and the UK, I wanted to follow up on the relative quality of genomic surveillance in the US and the UK. 1/12
First thing to clarify, in the @nytopinion opinion piece yesterday (nytimes.com/2020/12/22/opi…), it's mentioned that "since Dec. 1, Britain has sequenced more than 3,700 coronavirus cases, compared with fewer than 40 cases in the United States, according to Trevor Bedford". 2/12
As of today, the UK has shared to @GISAID 23,377 genomes during Dec and the US has shared 8033 genomes. However, the UK turnaround time has been much faster with 5010 specimens that were collected in Dec shared vs 65 collected in Dec and shared by the US. 3/12
Read 12 tweets
22 Dec 20
Following up on general thoughts on antigenic drift of #COVID19 from this weekend, I wanted to discuss what we know about the new variant of SARS-CoV-2 thats emerged in the UK. 1/17
This variant is referred to as the B.1.1.7 lineage in cov-lineages.org nomenclature and clade 20B/501Y.V1 in @nextstrain nomenclature and can be seen here within circulating viral diversity, where the variant lineage is highlighted in orange (nextstrain.org/ncov/europe?c=…). 2/17
Broadly, I'd characterize the source of concern as arising from the combination of:
1. Multiple mutations that from sequence composition alone are suggestive of biological importance
2. Observed rapid epidemic spread
3/17
Read 17 tweets
19 Dec 20
With #COVID19 vaccine efficacy of ~95%, I'm looking forward to vaccine distribution in 2021 bringing the pandemic under control. However, I'm concerned that we'll see antigenic drift of SARS-CoV-2 and may need to update the strain used in the vaccine with some regularity. 1/18
First, some background. RNA viruses all evolve extremely rapidly, but some like influenza are able to accept mutations to their surface proteins in such a way that they can partially escape human immunity. This process is known as "antigenic drift". 2/18
For influenza, this necessitates regular vaccine updates to keep up with an evolving virus population. Other RNA viruses like measles mutate quickly but are unable to change protein structure to escape from immunity and so these vaccines don't need updating. 3/18
Read 18 tweets
17 Dec 20
There has been a significant question about the degree to which Thanksgiving holiday and associated travel and social gatherings may have contributed to transmission of #COVID19. Here I try to briefly address this question. 1/8
Based on known incubation periods (nejm.org/doi/full/10.10…), we expect, on one end, some infections arising on Nov 26 to become symptomatic on Nov 30 and on the other end, for some infections arising on Nov 30 to become symptomatic on Dec 6. 2/8
This brackets the window where we expect most of the increased case load to be. However, most states only list cases based on date of report rather than date the case became symptomatic. This causes jitter that's hard to deal with when looking for a Thanksgiving effect. 3/8
Read 8 tweets
12 Dec 20
Although the US is continuing to hit records for daily #COVID19 cases reported, the rate of exponential growth has slowed. Mortality is still catching up to increased case loads and I expect daily deaths reported to further increase. 1/8
This plot summarizes the overall picture. Bubble size is proportional to daily cases per capita from @COVID19Tracking and bubble color shows Rt from rt.live. Timepoints are shown up to two weeks ago due to delay in reliable estimates of Rt. 2/8
The Midwest and Mountain West had rapid growth during October resulting in large epidemics in November, but they're now starting to plateau or decline in incidence. Although current incidence is lower, the epidemic is still growing in much of the East Coast. 3/8
Read 8 tweets
10 Dec 20
The US reported over 3000 deaths from #COVID19 today and the 7-day average of deaths has hit a record with today's average of 2276. Here, I dig into these grim mortality numbers and look at deaths across ages and across weeks in the epidemic. 1/14
I'm using data from @CDCgov (cdc.gov/nchs/nvss/vsrr…) that records weekly deaths involving COVID-19 as well as deaths from all causes. These data use actual date of death but there is a reporting lag. 2/14
CDC reports 261k deaths involving COVID-19 in this dataset. Over half of these deaths are in individuals 75 or older and over three quarters are in individuals 65 or older. 3/14
Read 14 tweets
4 Dec 20
The US is reporting over 2000 deaths per day from Dec 1 and I believe will do so consistently throughout December based on daily case loads above 120k starting early November. 1/4
A drop in reporting over Thanksgiving weekend has made for some difficulty in directly comparing 7-day averaged deaths, but the trend is clear. Red bars are daily reported deaths from @COVID19Tracking and black line is 7-day sliding average. 2/4
The simple projection of 1.7% of reported cases into deaths 22 days later has remained largely accurate, although drop of reporting during Thanksgiving weekend is quite clear. We'll know soon whether 7-day average returns back to projection. 3/4
Read 4 tweets