Update: the variants vs. the vaccines

By Juan Cambeiro, Analyst at Metaculus

In February 2021 we provided an update on COVID-19, outlining how the new variants would shape this next phase of the pandemic. Indeed, the story of the pandemic since then has largely been of a race between vaccinations and the spread of worrisome variants.

We have learned a lot about the variants since February, and relevant forecasts on Metaculus have been updated by our community accordingly. Overall, the information that has emerged paints a mostly optimistic picture: most of our vaccines induce robust and broad immune responses and work well against most of the variants we know about. Barring an unwelcome development like the emergence of a true vaccine-evading variant — of which there is an 11% chance between now and April 2022 — the end of the pandemic is in sight.

The goal of this update is to pull together much of the recent information on the new, known variants — the risks they pose, which warrant the most concern and what we know about them — and forecast what the landscape of variants might look like in the coming months. For more background on this topic and actionable steps that can be taken in response (which remain largely the same), check out the original Medium post.

SARS-CoV-2 continues to undergo rapid evolution

In February we noted that on a global level, SARS-CoV-2 — the virus that causes COVID-19 — is currently mutating at a rate even quicker than the fastest evolving seasonal influenza virus. This remains the case — with ~2.7 substitutions/year in the spike protein compared to ~2.5 substitutions/year in hemagglutinin for seasonal influenza virus A (hemagglutinin can be roughly thought of as flu’s equivalent of the SARS-CoV-2 spike protein).

We don’t yet know if SARS-CoV-2 will continue to sustain this rate of mutation. It might be especially high only for the time being because the virus is still adapting to transmit efficiently among humans. The rate of mutation being down slightly from earlier this year — going from ~2.9 substitutions/year to ~2.7 substitutions/year — supports this scenario. However, it’s also possible that partially immune populations may push continued evolution of the spike protein toward evading population immunity, much like the scenario that plays out with seasonal flu.

However, mutation rates are not the whole story. Measles and influenza, for instance, both have similarly high mutation rates, but while the former has a vaccine that has continuously protected against it since 1968, the latter requires yearly boosters. Evidently, influenza evolves in a way to evade protective immunity in a way that measles does not. It seems probable that a polyclonal antibody response that is broadly focused is the key factor at play — i.e., measles vaccination induces a wide range of antibodies with many targets, while flu vaccination induces a narrow range of antibodies with just a few targets (meaning if even a single target is mutated, immunity might be eroded substantially). COVID vaccination seems to fall somewhere in between measles and influenza vaccination, though there is significant variation depending on the type of vaccine used — and the variants being protected against.

Current mutation rate of SARS-CoV-2, Trevor Bedford

The most important variants to track

In February, we discussed three variants of concern (VOCs): B.1.1.7 (first detected in the UK), B.1.351 (first detected in South Africa), and P.1 (first detected in Brazil). However, additional VOCs, VOIs, and VUIs have emerged since February and warrant further study and focused attention, including B.1.427/B.1.429 (first detected in California), B.1.526 (first detected in New York), P.2 (first detected in Brazil), and B.1.617 (first detected in India). Another new VUI that we have extremely limited data on is C.37 (first detected in Peru). All of these have mutations that are concerning in one way or another. We’ll dig into this next.

Key mutations of the variants

First, let’s summarize some of the key spike mutations that are common to at least two of the variants discussed here. (Note that of the variants discussed here all have the D614G mutation, which became completely globally dominant early on the pandemic and is more transmissible — I won’t discuss it beyond this). Generally, the mutations are either thought to result in an increase in transmissibility or increase immune escape. Of the variants we currently know about, B.1.1.7, B.1.351, and P.1 all share the N501Y mutation — this mutation likely increases transmissibility. B.1.351, P.1, B.1.526, P.2, and a subset of B.1.617 also have the E484K/E484Q mutation, which is probably the single most important location for partial immune escape. Another very important mutation for partial immune escape is K417N. A somewhat important mutation for partial immune escape is L452R.

Juan’s table 1. Credit to covariants.org, outbreak.info, Jesse Bloom, Kristian Andersen

Something to keep in mind is that while in theory a combination of these mutations might result in a new, even more worrisome variant, in reality the different mutations might instead result in a less robust overall spike structure. It’s difficult to predict the effect of different combinations of mutations. Indeed, an interesting case is B.1.1.7 carrying E484K, which in theory might be a fearsome new variant but in reality has not really taken off yet.

What we know about the variants

Next, let’s summarize what we know about the most important of the new variants (except C.37 since we know next to nothing about it) and the original three VOCs, which I’ve displayed in a table below. Of the variants that have been detected and at least somewhat studied, the ones in this table I’ve created are the most concerning. Where applicable, I’ve used estimates by the most reliable study on said variant.

Juan’s table 2. Credit to covariants.org, outbreak.info, cov-lineages.org, Davies et al., Davies et al., Pearson et al., Faria et al., Naveca et al., Deng et al., Funk et al. For the studies on immune evasion, see the next section.

This table is pretty complex and chock full of information, so I’ll note a few things that stand out:

• B.1.1.7 is by far the most well-characterized variant. We now know it’s almost certainly significantly more transmissible (~55%) and more lethal (~61%). Per sequencing data, it’s also probably more prevalent worldwide than all the other variants in the table combined. Fortunately, it is not able to substantially evade immunity.
• B.1.1.7, B.1.351, and P.1 have all become very dominant (>75% prevalence) in the countries where they were first identified (respectively: United Kingdom, South Africa, and Brazil). All three share the N501Y mutation that likely increases transmissibility.
• B.1.351, P.1, B.1.526, P.2, and a subset of B.1.617 all share the E484 mutation — which, to our knowledge, is the single most relevant mutation for immune escape
• B.1.617 — which according to media reports is behind the ongoing devastating surge of cases in India — does not, on the face of it, seem to be that worrisome. Only a minority of sequenced cases have the E484 mutation. Moreover, the L452R mutation is thought to only have a modest effect on immune evasion. Still, B.1.617 is the least well-characterized variant in the table, and more information may yet emerge that this variant is a major factor behind the surge of cases in India.

Likelihood that SARS-CoV-2 will adapt to evade vaccine-induced immunity

As we’ve seen when looking at specific variants, it’s already well-established that some variants are more transmissible and/or can partially evade vaccine-induced immunity. Indeed, the Metaculus community has ongoing questions that touch on these topics:

• Transmissibility: 90% chance (slightly down from 94% in February) that a variant which is >30% transmissible will infect >10 million people before mid-2021 and 65% chance (slightly down from 69% in February) that a variant which is >50% transmissible will infect >10 million people before mid-2021.

• Partial immune evasion: 55% chance (unchanged from 54% in February) that a new variant will significantly (>=50% difference in the attack rate in seropositive subjects) evade the immunity of previously infected people before 2022, and a 75% chance (somewhat up from 70% in February) that the U.S. CDC will recommend before 2023 that previously vaccinated people be revaccinated to protect against new variant(s).

A key remaining uncertainty is whether a variant will emerge that can result in a significant reduction in the real-world effectiveness of our best vaccines (no such variant has yet emerged).This would necessitate substantially evading antibodies induced by vaccination. The Metaculus community has assigned an 11% probability to such a variant of high consequence emerging before April 2022.

U.S. CDC definition of a variant of high consequence

There are other variant classifications, including variant of concern (VOC) and variant of interest (VOI). For our purposes, let’s add a variant under investigation category (VUI) for variants a step below VOI, i.e., for possible rather than predicted worrisome features.

• VOC: B.1.1.7, B.1.351, P.1, B.1.427/B.1.429
• VOI: B.1.526, P.2
• VUI: B.1.617

U.S. CDC variant classifications

On the whole, relevant Metaculus forecasts illustrate a largely unchanged picture with respect to the transmissibility and partial immune evasion properties of variants that have emerged and have yet to emerge. This is mostly good news, as this involves a flattening of fat tails for worst-case scenarios since data that has come in since February has failed to substantially shift the overall picture in either direction (of course, this also unfortunately means that the best-case scenarios of no variants arising that are more transmissible/can partially evade immunity can be almost completely ruled out). Our new question that directly gets at the question that is probably of most relevance at the moment — whether a variant of high consequence will emerge — indicates a one-in-nine chance of the worst-case scenario coming to pass in the next year.

The general protective effect of vaccines against a rapidly changing SARS-CoV-2

The single most important factor to prevent immune escape by variants in already-vaccinated people is, as mentioned previously, for the vaccines to have induced a broadly focused polyclonal antibody response. One way to assess this is to look at antibody levels induced against worrisome variants in lab experiments. Another way is to analyze vaccine trial data/real-world vaccine efficacy data against specific variants. We’ll discuss both of these later. But without even having to know anything about which variants are spreading, we can examine (1) the extent to which different vaccines induce broadly focused polyclonal antibody responses, (2) T-cell responses, and (3) the general results of phase III trials that do not specifically break down efficacy against variants.

(1) Broadly focused polyclonal antibody response: With respect to generating a very broad antibody response, the mRNA vaccines (Pfizer/BioNTech and Moderna) seem to be the most impressive. This appears to be in large part because both Pfizer/BioNTech and Moderna code for a prefusion-stabilized spike protein (antibodies can more readily bind to this locked conformation), which seems to induce a good polyclonal response. Other non-mRNA vaccines that use this prefusion-stabilized spike protein include the Johnson & Johnson, Novavax, and CureVac vaccines.non-mRNA vaccines that do not code for the prefusion-stabilized spike — including AstraZeneca/Oxford, Sputnik V, and Sinovac — probably do not induce as robust of a broad antibody response.

Credit: Bing Chen of Harvard Medical School

(2) T-cell response: Another key aspect of the immune response (which is often overlooked) is the importance of T-cells. Notably, it appears antibody levels are not predictive of T-cell memory (meaning one can have low antibody levels but robust T-cell memory) and asymptomatic infections result in levels of T-cell memory similar to those induced by symptomatic infections. Most people develop a robust T-cell response just 10 days after their first vaccine dose, and the T-cell response after one dose in previously-infected individuals is comparable to naive individuals receiving two vaccine doses. Crucially, T-cell responses are largely unaffected by key variants of concern. All of the vaccines that have undergone phase III trials seem to induce great T-cell responses. Interestingly, the Astra/ZenecaOxford vaccine seems to induce a better T-cell response than the mRNA vaccines.

Robust T-cell response against multiple variants, Sette et. al

(3) General results of phase III trials: Phase III trial vaccine efficacy results against symptomatic disease have generally been very encouraging, and the extent of protection against severe disease is even more encouraging (nearing 100% for some vaccines). However, there is the notable caveat that some of these results — including for Pfizer, Moderna, Sputnik V, and AstraZeneca — came out before worrisome variants were detected. The Economist has a nice graphic summary of results on protection against symptomatic disease:

Select phase III efficacy results, The Economist

Vaccines vs. variants

Finally, let’s take a look at how well the vaccines work against the specific variants discussed here. For this, let’s consider both vaccine efficacy against disease caused by variants according to trials/real-world data and antibody levels against variants according to lab studies.

Juan’s table 3. Credit to Pfizer, Moderna, Johnson & Johnson, AstraZeneca, Novavax, Chen et al., Karim & Oliveira, Planas et al., Hoffman et al., Shen et al., Wu et al., Liu et al., Munitz et al., Edara et al., Wang et al., Dejnirattisai et al., McCallum et al., Deng et al., West et al., Zhou et al.

The first thing to note is just how much blank space there is.We still need a lot more data! In fact, we’re missing even more than is apparent here because I’ve limited the vaccines included in the table to those that have been the most well-studied. Still, a few things are clear:

• With the possible exception of AstraZeneca/Oxford vaccine efficacy data against B.1.351, there is no truly worrisome result here. It’s increasingly evident that the vaccines listed here work quite well. Still, it’s worth keeping an eye out for further studies on B.1.351, P.1, B.1.526, and P.2 — these four warrant the most concern with respect to immune escape. And of these worrisome four, B.1.351 and P.1 are the most concerning (B.1.526 is largely limited to the New York City area, where vaccines are being rolled out and where cases are falling; P.2 is being replaced by P.1 in parts of Brazil).
• B.1.1.7 and B.1.427/B.1.429 warrant little concern with respect to immune escape. This is seemingly backed up by the fact that cases in the UK (where B.1.1.7 was first detected) and California (where B.1.427/B.1.429) remain very low (both the UK and California are rolling out vaccines at a brisk pace).
• We know next to nothing about vaccine efficacy/antibody levels against B.1.617. We need focused attention on characterizing it. Still, data from table 2 suggests that it is likely not the main factor at play in India’s current surge of cases.

Takeaway

The key takeaway is that recent data indicate that most vaccines protect very well against the variants we know about, and the end of the pandemic is possible in the near future unless a variant that is truly vaccine-evading emerges — of which the Metaculus community thinks there is an 11% chance between now and April 2022. The partial immune escape by existing variants and the high likelihood of further variants emerging that can partially escape immunity likely means we will need booster shots — with a 75% chance the U.S. CDC will recommend this before 2023. But even without booster shots immunity against severe disease is likely to be retained, though mild vaccine breakthrough cases might occur (currently, vaccine breakthrough cases are pretty rare).

Since February, significant progress has been made in ramping up genomic sequencing, enabling us to have a much clearer picture of the global landscape of variants. Still, more is needed, especially in the Global South. Though much has been learned since February, more also remains to be learned about vaccine efficacy against specific variants and more neutralization lab experiments are needed (side note: a small number of high-quality studies is generally more valuable than a plethora of low-quality ones!).

At the very least, this much is clear: vaccinating as many people as fast possible — even if this means delaying second doses for countries in early stages of vaccine rollout (and, once second dose administrations begin, prioritizing them for people over 60 years old and deprioritizing them for those previously infected) — remains the single best way to both reduce the number of infections/deaths and prevent the emergence of a variant of high consequence (i.e., a true vaccine-evading variant). When combined with sensible non-pharmaceutical interventions to blunt case surges, a path out of the pandemic can be realized.

Follow Juan Cambeiro on Twitter @juan_cambeiro and Metaculus @metaculus