Transmissibility of MERS-CoV in camel populations
Force of infection (FoI)
To estimate the FoI, we fitted four different models of seroconversion to age-stratified seroprevalence data extracted in our previous systematic review of MERS-CoV in camels8, assuming the seroprevalence data was beta-binomially distributed. We found that allowing a proportion of calves to be born with protective maternally acquired antibodies (mAbs) in model 3 improved model fit, as did the inclusion of seroreversion due to waning of antibodies acquired following infection in model 2, though the additional improvement in fit afforded by seroreversion on top of mAbs in the best fitting model (model 4) was fairly small (Table S1). We assumed test sensitivity and specificity were high for both neutralisation and non-neutralisation-based tests. The ranking of model fit was robust to the use of alternative assumptions where sensitivity of neutralisation tests was modelled to be lower at 85% (Table S2). Model selection was also generally robust to exclusion of each dataset from the analysis, with the model with mAbs alone or with mAbs and seroreversion outperforming the others (Figure S2).
Parameters governing rates of antibody waning and data overdispersion were estimated as common to all studies, while the FoI was allowed to be study-specific. We estimated that mAbs waned rapidly in the first few months of life, lasting on average 2 months (95% credible interval (CrI): 1–4 months) in our best fitting model, but that antibodies wane slowly following infection, lasting approximately 17 years (95% CrI: 9–33 years) – approaching the lifespan of camels. However, it can be difficult to distinguish life-long antibodies from repeated boosting using catalytic models. Parameter estimates were largely robust to exclusion of each data set with the exception of the data collected in Egypt which, when excluded, increased the estimated duration of mAbs to 4.8 months, similar to the 4.2 months estimated in model 3 (Figure S3). The overdispersion parameter, k, was estimated to be 2.5 (95% CrI: 2.0, 3.2), meaning the variance of the data was estimated to be around 3.5 times greater than what would be expected if the data were binomially distributed. When an uninformative prior for k was used, the model tended to maximise k meaning an extreme level of overdispersion could on its own account for the patterns observed in the data, irrespective of other epidemiological parameter values which were then unidentifiable. To circumvent this issue, a half normal prior with a standard deviation of 0.5 was used to constrain k to values we believe to be more plausible.
The annual FoI of MERS-CoV in camels was generally higher in populations sampled in the Middle East, and lower in those sampled in South Asia and Africa (Fig. 1, Table 1). This trend was consistent across all four models (Table S3). The posterior mode for the FoI ranged from 0.1-3.0 across most study populations, except for in the population in UAE where seroprevalence was very high (> 85%) in both calves and adults, perhaps indicative of a recent outbreak, and the FoI was estimated to be 7.1/year. Such high FoI estimates are necessary to explain the high seroprevalence measured in younger animals when assuming endemic transmission, but it is important to note that there is a risk of over-estimation of FoI for populations with seroprevalence approaching 100% since all high FoI values fit the data equally well. This effect is likely reflected by the long tails of some of the posterior distributions presented in Fig. 1A; we therefore chose to present the posterior mode as the central FoI estimates as we believe this to be a more representative than the mean. The best fitting model matched the data well, with model estimates overlapping the age-stratified seroprevalence data with only a few exceptions (Fig. 1B). The very high levels of seroprevalence measured in young calves in Tunisia and Kingdom of Saudi Arabia (KSA)16 were underestimated, and the model struggled to reproduce the data collected in one study in KSA where seroprevalence was very high in calves and dropped considerably in adults17, which was not seen in any other study.
Table 1
Population specific estimates of the transmissibility of MERS-CoV in camels.
|
Dataset
|
Seroprevalence (%)
|
Test*
|
FoI, l
Model 4:
|
R0
|
relative infectiousness of reinfections
|
1%
|
50%
|
Africa
|
Egypt18
|
< 2yrs 37% (n = 595)
≥ 2yrs 82% (n = 1946)
|
MN
|
0.5 (0.4, 0.7)
|
4.2 (3.7, 5.1)
|
1.9 (1.8, 2.1)
|
Egypt16
|
< 2yrs 16% (n = 447)
≥ 2yrs 84% (n = 1586)
|
MN
|
0.5 (0.4, 0.6)
|
4.0 (3.5, 4.7)
|
1.9 (1.8, 2.0)
|
Ethiopia19
|
1- ≤2yrs 93% (n = 31)
2-13yrs 97% (n = 157)
|
PM
|
2.7 (1.6, 9.7)
|
14.9 (9.6, 44.0)
|
3.0 (2.6, 4.9)
|
Kenya20
|
< 6m 39% (n = 61)
6m-2yrs 21% (n = 80)
> 2yrs 61% (n = 194)
|
PM
|
0.2 (0.2, 0.4)
|
2.6 (2.1, 3.5)
|
1.7 (1.4, 1.8)
|
Kenya21
|
1-4yrs 73% (n = 285)
4-6yrs 98% (n = 116)
6yrs 98% (n = 476)
|
ELISA
|
1.0 (0.8, 2.3)
|
6.8 (5.5, 13.4)
|
2.3 (2.2, 2.9)
|
Kenya22
|
< 4yrs 36% (n = 319)
> 4 < 7yrs 59% (n = 70)
> 7yrs 82% (n = 760)
|
ELISA
|
0.3 (0.2, 0.4)
|
2.9 (2.5, 3.7)
|
1.7 (1.6, 1.9)
|
Senegal16
|
< 2yrs 29% (n = 17)
≥ 2yrs 69% (n = 181)
|
MN
|
0.3 (0.2, 0.6)
|
2.9 (2.3, 4.6)
|
1.7 (1.5, 2.0)
|
Tunisia16
|
< 2yrs 100% (n = 28)
≥ 2yrs 87% (n = 754)
|
MN
|
0.8 (0.6, 2.2)
|
5.8 (4.6, 12.8)
|
2.2 (2.0, 2.8)
|
Tunisia19
|
< 2yrs 30% (n = 46)
≥ 2yrs 54% (n = 158)
|
PM
|
0.2 (0.1, 0.3)
|
2.2 (1.8, 3.1)
|
1.5 (1.3, 1.7)
|
Uganda16
|
< 2yrs 52% (n = 150)
≥ 2yrs 66% (n = 350)
|
MN
|
0.3 (0.2, 0.6)
|
3.2 (2.6, 4.5)
|
1.8 (1.7, 2.0)
|
Middle East
|
Iraq16
|
< 2yrs 33% (n = 6)
≥ 2yrs 57% (n = 21)
|
MN
|
0.2 (0.1, 7.5)
|
2.6 (1.7, 35.3)
|
1.7 (1.3, 4.3)
|
Iraq23
|
< 2yrs 89% (n = 44)
2-4yrs 81% (n = 58)
> 4yrs 86% (n = 78)
|
ELISA
|
1.8 (0.9, 9.4)
|
10.6 (6.4, 43.0)
|
2.7 (2.3, 4.8)
|
Jordan16
|
< 2yrs 50% (n = 82)
≥ 2yrs 92% (n = 222)
|
MN
|
1.0 (0.7, 2.4)
|
6.9 (5.0, 13.8)
|
2.3 (2.1, 2.9)
|
Jordan24
|
≤ 2yrs 74%a (n = 31a)
> 2yrs 100%a (n = 14a)
|
ELISA
|
2.6 (1.2, 9.5)
|
14.8 (7.8, 43.1)
|
3.0 (2.4, 4.8)
|
KSA25 1992–2010
|
≤ 2yrs 55% (n = 104)
> 2yrs 95% (n = 98)
|
ELISA
|
1.8 (1.1, 7.8)
|
10.9 (7.2, 36.3)
|
2.7 (2.4, 4.4)
|
KSA25 2013
|
≤ 2yrs 73% (n = 77)
> 2yrs 93% (n = 187)
|
ELISA
|
1.2 (0.7, 2.8)
|
7.5 (5.3, 15.7)
|
2.4 (2.1, 3.0)
|
KSA17
|
1-2yrs 93% (n = 71)
3-5yrs 78% (n = 100)
|
ELISA
|
2.3 (1.2, 9.5)
|
13.2 (7.5, 43.1)
|
2.9 (2.4, 4.8)
|
KSA26
|
< 1 year 72% (n = 65)
1-3yrs 95% (n = 106)
4-5yrs 97% (n = 76) >5yrs 92% (n = 63)
|
ppNT
|
3.0 (1.7, 9.1)
|
16.6 (10.4, 41.7)
|
3.1 (2.6, 4.7)
|
KSA16
|
< 2yrs 82% (n = 11)
≥ 2yrs 82% (n = 211)
|
MN
|
0.6 (0.4, 2.2)
|
4.5 (3.4, 12.6)
|
2.0 (1.8, 2.8)
|
UAE11
|
≤ 1 year 85% (n = 108)
2-4yrs 97% (n = 340)
> 4yrs 96% (n = 310)
|
ELISA
|
7.1 (3.5, 9.8)
|
33.7 (18.5, 44.5)
|
4.2 (3.2, 4.9)
|
South Asia
|
Bangladesh27
|
< 2yrs 9% (n = 11)
≥ 2yrs 36% (n = 44)
|
ppNT
|
0.1 (0.0, 0.3)
|
1.7 (1.3, 3.2)
|
1.3 (1.1, 1.8)
|
Pakistan28
|
≤ 2yrs 29% (n = 89)
2.1-5yrs 30% (n = 208)
5.1-10yrs 51% (n = 180)
> 10yrs 49% (n = 88)
|
ELISA
|
0.1 (0.1, 0.2)
|
1.9 (1.7, 2.3)
|
1.4 (1.3, 1.5)
|
Pakistan29
|
≤ 3yrs 58% (n = 177)
3.1-10yrs 79% (n = 712)
> 10yrs 81% (n = 161)
|
ELISA then MN
|
0.5 (0.4, 0.7)
|
3.9 (3.3, 4.9)
|
1.9 (1.8, 2.1)
|
Global
|
Rate of waning mAbs, w
|
|
|
4.9 (3.2, 9.6)
|
|
|
Rate of waning Abs, s
|
|
|
0.06 (0.03, 0.11)
|
|
|
Overdispersion, k
|
|
|
2.5 (2.0, 3.2)
|
|
|
*MN = micro neutralisation test, ppNT = pseudo particle neutralisation test, PM = protein micro-array, ELISA = Enzyme linked immunosorbent Assay.
Basic reproduction number
Basic reproduction number (R0)
R 0 estimates can provide a more widely used intuitive measure of transmissibility which are more readily comparable with other diseases. We translated FoI into R0 using a dynamic, age-stratified, stochastic model of MERS-CoV transmission (please see Methods for a detailed description of model assumptions). Estimates were sensitive to the assumed relative infectiousness of reinfected animals (Table 1). When reinfected animals were assumed to be 1% as infectious as animals experiencing a primary infection (based on viral shedding data from the control arm of the ChAdOx vaccine field study6), central R0 estimates ranged from 3 to 34 in the Middle East, compared with 2 to 15 in populations sampled in Africa and 2 to 4 in South Asia. As a sensitivity analysis, when infectiousness was assumed to be only halved in reinfections (based on assuming a logarithmic rather than linear relationship between the measured viral shedding and infectiousness), R0 ranged from 2 to 4 in the Middle East, 1 to 3 in Africa, and 1 to 2 in South Asia. R0 estimates for populations with very high FoI estimates were most sensitive to assumptions about immunity as, in these populations, a higher proportion of infections at endemic equilibrium are reinfections and are therefore affected by relative infectiousness parameters. Neither varying the duration of complete immunity following infection, nor the relative susceptibility of previously infected individuals had a considerable effect on R0 estimates (Table S4). When using estimates for the longer duration of mAbs (4.2 months) and for the FoI from the second-best fitting model without seroreversion, R0 estimates were similar to those using best fitting model with seroreversion, albeit slightly lower (Table S4).
The critical community size (CCS)
We used the transmission model to estimate the population size above which the extinction of MERS-CoV transmission by chance becomes unlikely - the critical community size (CCS)30. We considered three different transmissibility levels spanning our R0 estimates across different settings: low (South Asia and parts of Africa), moderate (Kenya and parts of Middle East), and high (Ethiopia and parts of Middle East) (Figure S4). The CCS varied between approximately 10,000–70,000 camels depending on the transmissibility, seasonality of calving and underlying herd structure assumed in the transmission model (Table 2, Fig. 2). The CCS decreased as transmission intensity increased, except for when births were highly seasonal, and the population was modelled as homogeneous. In this case, high transmissibility resulted in a larger CCS than low or moderate transmissibility. This likely represents a complex interaction between seasonality, transmissibility, and accumulation of susceptible individuals. Transmission could be sustained in smaller populations when births were less seasonally forced, and when the population was assumed to be homogenous as opposed to being structured into weakly connected patches intended to represent large herds or communities. Under our alternative assumption that viral shedding is proportional to the log of infectiousness meaning past infection reduces infectiousness by only 50%, the CCS was smaller, with only 1,000–30,000 camels needed to sustain transmission across depending on the transmission setting (Table 2, Fig. 2).
Table 2. The estimated critical community size of MERS-CoV in camels under two alternative values for the relative infectiousness of reinfected animals (rinf), in different transmission settings (R0). Estimates are shown for models assuming a homogeneous population of perfectly mixed animals and a structured population in which animals have more contact with those in their “patch” and weaker contact with those in surrounding patches. Estimates are shown assuming births follow seasonal patterns as strong as those observed in KSA (d = 1) and alternatively with a weaker seasonality (d = 0.5).
Periodicity of transmission
When births are assumed to follow seasonal patterns representative of those observed in KSA (see Methods), the number of infections over time has an annual periodicity in large populations, with peaks of a similar size occurring each year (Fig. 3.A). In small populations, or when transmissibility is low, reflecting estimates for camel populations in South Asia and parts of Africa, biennial, triennial, and quadrennial periodicities - with patterns in the magnitude of annual peaks in infections repeating over 2-, 3- or 4-year cycles – are detected based on autocorrelation coefficients in a proportion of stochastic iterations (Fig. 3.B). No seasonality in infections is observed when births are non-seasonal.
The impact of vaccination
Optimal target age
We extended the transmission model to simulate the impact of age-targeted vaccination under multiple efficacy scenarios based on RNA shedding data from field studies and remaining uncertainties (Methods). The optimal target age for routine vaccination was assessed in a large population of camels so that overarching trends were not obscured by stochasticity. In our conservative scenario (scenario 1), in which vaccination reduces infectiousness of subsequent infections in all vaccinated animals but had no effect on susceptibility, vaccination led to the greatest reduction in infection incidence when calves were targeted in the first few months of life (Fig. 4A). In the absence of vaccination, most animals were first infected when they were < 1-yr old across all modelled transmission settings. In targeting younger calves, vaccination precedes first infection in a greater number of individuals, reducing their subsequent infectiousness. The reduction in incidence afforded by targeting younger animals was larger in higher transmission settings where first infections occurred earlier, with reductions in incidence diminishing quicker as the target age class was increased compared to in lower transmission settings. When the duration of vaccine induced effects was assumed to be relatively short and transmission intensity was moderate or high, vaccinating very young calves shifted the average time to first infection into older age groups, leading to a small increase in the annual incidence in adult animals of up to 10 per 1000 animals under our core model assumptions (Figure S6). Vaccinating at 6 months allowed large reductions in overall incidence of infection without seeing considerable shifting of first infections into adult animals. Note that adult incidence is considered here given it may be a proxy for zoonotic spillover risk to humans.
Under our optimistic scenario (scenario 2) in which we assumed vaccination reduced both infectiousness and susceptibility in all vaccinated animals, we saw the same pattern as in scenario 1, with greatest impact achieved by vaccinating in the first few months of life, and no notable increase in adult incidence when targeting 6-month-olds. Vaccinating older age groups after most first infections had occurred had almost no effect on incidence in scenario 1, whereas a slight reduction was still achieved by reducing the susceptibility of the older animals to reinfection in scenario 2 (Figure S7). In our third scenario, in which vaccination was only effective as a booster for previously infected animals with no impact in naïve animals, the optimal age for vaccination was early adulthood but even then, the reduction in incidence was minimal at < 8% across all transmission intensities (Figure S7). The optimal target age for vaccination was robust to our different assumptions about the relationship between viral load and infectiousness.
The impact of vaccination on transmission
To explore the characteristics of the MERS-CoV vaccine and the vaccine coverage that would be necessary to achieve considerable reductions in infection incidence among camels, we simulated the impact of vaccination of 6-month-old calves in two modelled populations: one of 2 million camels comparable in size to that of KSA; and one of 75,000 camels comparable to that of a small camel-keeping Kenyan county.
In a population of 2 million camels divided into large homogenous patches, assuming the vaccine reduces infectiousness but not susceptibility, the vaccination coverage required to half the total incidence over the 10 years following introduction was between 50–90% in 6-month-olds, depending on the duration of vaccine induced effects and the transmission intensity (Fig. 4B). When vaccine induced effects were long lasting, 50% coverage was required to half incidence in low transmission intensity settings, rising to approximately 80% coverage needed in high transmission intensity settings (Fig. 4B). When effects lasted 3 years, coverage of 70% was needed in low transmission settings rising to 90% when transmission intensity was high, and when vaccine induced effects only lasted one year, incidence could not be halved under any modelled setting. Alternatively, in a population of 75,000 camels, stochastic effects amplified the impact of vaccination: a coverage of < = 50% in 6-month-olds could half total incidence in the 10 years following vaccine in low transmission intensity settings, even if effects only lasted 1 year. In a moderate transmission intensity setting, between 50–70% coverage was needed, and in high transmission intensity settings 70–90% coverage, depending on duration of vaccine induced effects. Across all transmission intensities, assuming the vaccine reduced susceptibility of vaccinated animals to 50% or 75% (efficacy scenario 2) only afforded a very small (~ 1% on average) additional reduction in incidence compared to when assuming the vaccine reduced infectiousness alone (Figure S8).
In the population of 2 million, when R0 was low, coverage was high, and the effects of the vaccine were long-lasting, vaccination was capable of interrupting transmission and led to stochastic fadeout. In these cases, the difference in incidence between stochastic runs was often large (the 2.5% and 97.5% quantiles are represented by transparent ribbons in Fig. 4B). In low and moderate intensity settings, vaccination was able to interrupt transmission when coverage in 6-month-olds was very high and the vaccine induced effects lasted at least 3 years (Table 3). In high transmission intensity settings transmission was only interrupted when coverage was 100% and vaccine induced effects lasted 10 years. In the smaller population of 75,000 divided into homogenous patches of 3,000, stochastic fadeout occurred at lower coverages and across a wider range of scenarios. Vaccination was capable of reliably interrupting transmission when coverage ranged from 40–80% depending on transmission intensity and duration of vaccine induced effects.
Table 3
The coverage necessary to interrupt transmission at the population level in two modelled populations
|
|
Vaccine coverage (%) needed to interrupt transmission in a population of:
|
R0
|
1/ρ
|
75,000 split into patches of 3,000
|
2,000,000 split into patches of 80,000
|
3.5
|
1
|
40
|
NA
|
3
|
40
|
90
|
10
|
40
|
70
|
7.0
|
1
|
80
|
NA
|
3
|
60
|
100
|
10
|
60
|
90
|
14.0
|
1
|
80
|
NA
|
3
|
70
|
NA
|
10
|
60
|
100
|