Surveillance of endemic human coronaviruses (HCoV-NL63, OC43 and 229E) associated with childhood pneumonia in Kilifi, Kenya

Introduction: Human coronaviruses (HCoVs) circulate endemically in human populations, often with seasonal variation. We describe the long-term patterns of paediatric disease associated with three of these viruses, HCoV-NL63, OC43 and 229E, in coastal Kenya. Methods: Continuous surveillance of pneumonia admissions was conducted at the Kilifi county hospital (KCH) located in the northern coastal region of Kenya. Children aged <5 years admitted to KCH with clinically defined syndromic severe or very severe pneumonia were recruited. Respiratory samples were taken and tested for 15 virus targets, using real-time polymerase chain reaction. Unadjusted odds ratios were used to estimate the association between demographic and clinical characteristics and HCoV positivity. Results: From 2007 to 2019, we observed 11,445 pneumonia admissions, of which 314 (3.9%) tested positive for at least one of the HCoV types surveyed in the study. There were 129 (41.1%) OC43, 99 (31.5%) 229E, 74 (23.6%) NL63 positive cases and 12 (3.8%) cases of HCoV to HCoV coinfection. Among HCoV positive cases, 47% (n=147) were coinfected with other respiratory virus pathogens. The majority of HCoV cases were among children aged <1 year (66%, n=208), though there was was no change in the proportion infected by age. HCoV-OC43 was predominant of the three HCoV types throughout the surveillance period. Evidence for seasonality was not identified. Conclusions: Overall, 4% of paediatric pneumonia admissions were associated with three endemic HCoVs, with a high proportion of cases co-occurring with another respiratory virus, no clear seasonal pattern, and with the age-distribution of cases following that of pneumonia admissions (i.e. highest in infants). These observations suggest, at most, a small severe disease contribution of endemic HCoVs in this tropical setting and offer insight into their potential future burden and epidemiological characteristics.


Introduction
To date, seven human coronaviruses (HCoVs) have been identified, of which four (NL63, HKU1, OC43 and 229E) are known to be endemic among humans [1][2][3][4] . Endemic HCoV types are detected in a small but non-negligible proportion of respiratory tract infections; mild cases occur across a wide age-range and severe disease is predominant in young children and the elderly [5][6][7][8] . A further three HCoVs have emerged in recent years and caused epidemics: SARS-CoV the agent of severe acute respiratory syndrome in China 9 , MERS-CoV the cause of Middle East respiratory syndrome in the Middle-East 10 and most recently SARS-CoV-2, the aetiological agent of the current pandemic of coronavirus disease 2019 (COVID-19) 11 . To date, there are limited preventive options against HCoV infections and no effective anti-viral treatment 12 . Understanding the epidemiology of HCoVs can play a critical role in prediction, prevention and control of HCoV infection. In addition, data on endemic HCoVs may inform expectations for SARS-CoV-2 if it becomes endemic. Our study aims to describe the circulation patterns of three endemic HCoVs (OC43, 229E, and NL63) over time using data from a long-term surveillance programme in a rural coastal setting in Kenya, 3° south of the equator.

Study setting
A prospective study was established in 2007 for long-term continuous respiratory virus surveillance among pneumonia admissions to Kilifi County Hospital (KCH) 6 in order to develop improved epidemiological understanding, estimate disease burden and provide suitable baseline data for future vaccine studies. KCH is the referral hospital within the Kilifi Health and Demographic Surveillance System (KHDSS), in the northern coastal region of Kenya 13 . The location experiences two rainy seasons approximately from April to July and October to December, with median maximum temperature of 33°C (IQR: 31-36), median minimum temperature of 23°C (IQR: [22][23][24], and median relative humidity of 78% (IQR: 71-87) (unpublished weather station data). Children aged 1 day to 59 months admitted to KCH with clinical symptoms of severe or very severe pneumonia were recruited. Written informed consent was sought from parents/guardians of the children prior to sample collection. In this paper we define severe pneumonia as history of cough or difficulty breathing and chest indrawing while very severe pneumonia is defined as history of cough or difficult breathing and at least one of inability to feed, prostration, unconsciousness, or oxygen saturation of <90% by fingertip pulse oximetry. We use the term pneumonia to refer to all cases of clinically severe or very severe pneumonia 14 . The variables extracted from the hospital surveillance database include; Demographic characteristics (sex, KHDSS residency status, age), presence/ absence of clinical features (history of cough, difficulty breathing, cyanosis, nasal flaring, chest indrawing, crackles, wheeze, inability to drink, vomits everything, fever defined as axillary temperature ≥37.5° C, oxygen saturation levels, conscious level: agitated, lethargic, prostration or unconscious, pneumonia status: severe or very severe), laboratory test results, and hospitalisation outcomes (admission to the high dependency unit, discharge outcomes: alive or dead)

Laboratory methods
Specimens collected between January 2007 and December 2019 were processed and screened for three HCoVs (OC43, NL63 and 229E) and at least 12 other respiratory viral pathogens (RSV (A and B), rhinovirus, HCoVs (NL63, OC43, 229E), influenza (A, B and C), parainfluenza virus (1-4), adenovirus, and human metapneumovirus) using real-time polymerase chain reaction (RT-PCR). Sample testing was initially performed in 2007 using the LightCycler Fast Start DNA MasterPLUS HybProbe kit (Roche) 6 , then multiplex RT-PCR using Qiagen Quantifast multiplex RT-PCR kit (Qiagen, United Kingdom) in triplex sets on an ABI 7500 system, from January 2007 until the present day 15,16 ; additionally, a proportion of samples were tested using a 33-pathogen multiplex quantitative PCR (FTD Resp-33, Fast Track Diagnostics, Sliema, Malta) as part of the multi-country PERCH study 17 , between August 2011 and December 2013. A variety of collection methods was used: nasal wash (2007 to 2009), nasopharyngeal flocked swab (2010 to 2014), or combined nasopharyngeal swab and oropharyngeal swab (2015 onwards).

Data analysis
Data analysis was done using STATA version 13.0 (Stata Corp, College Station Texas, USA). Summary statistics (counts, proportions, measures of central tendency and variation) are presented for continuous and categorical data as appropriate. We estimated unadjusted odds ratios to measure the association between demographic and clinical characteristics of the study participants and testing positive for HCoV. Three Poisson regression models, one for each HCoV type, were used to investigate the presence of seasonality. In the models a trend variable was included and residuals plotted against month. Identification of a strong pattern by visual inspection of the residual plots would suggest presence of seasonality. The chi-square test of proportional trends was used to test for a linear trend in the proportions of samples tested or not tested for HCoV over time. To check for an association between categorical variables the chi-square test of association or Fisher's exact test was used as appropriate. The analysis code is provided as Extended data 18 .

Ethical approval
This study was approved by the Kenya Medical Research Institute Scientific Ethics Review Unit (Approval number: KEMRI/SERU/CGMR-C/027/3178).

Characteristics of patients infected with HCoVs
During the 13 years of surveillance, there were 49,409 paediatric admissions of children aged 0-59 months at KCH. A total of 11,445 (23.2%) admissions were due to severe (n= 7808, 68%) and very severe (n=3637, 32%) pneumonia. Out of the eligible cases, 69.5% (n=7957) were tested for the three HCoVs while the remainder were not tested due to refusal of consent (13.8%), discharge (13%) or death (3.7%) prior to sample collection. Cases untested did not differ from those tested in age distribution or sex ratio, but were more likely to be very severe (40.0% versus 28.3%, Fishers exact P-value <0.001).
The characteristics of the patients positive for any and for each HCoV type or infection combination are described in Table 1. HCoV positive cases were predominantly children aged <1 year (66.2%) and those aged 12-23 months (18.2%). The burden of infection with at least one HCoV, among all pneumonia admissions, was highest in infants and decreased with increasing age (2.6% for those under 1 year and 0.7% for 12-23 months). The same pattern was seen for each individual HCoV type (not shown). However, the proportion of samples testing positive for HCoV (3.9%) did not vary with age group (Fisher's exact p-value = 0.753). Among all HCoV positive participants, mean age was 11 months (median 7 months), there were fewer females than males and fewer with very severe compared to severe pneumonia. At least half of the HCoV positives presented with fever (58%) and nasal flaring (53%).
Clinical outcomes of HCoV-infected patients Over a quarter of those positive for at least one of the three endemic HCoVs investigated in this study were admitted to the high dependency unit and of those positive for OC43, 13.5% (n=10) died while 7.8% (n=10) and 7.1% (n=7) died of those positive for NL63 and 229E, respectively. A large proportion of these deaths were observed among those with underlying co-morbidities ( Figure 1). None of the HCoVs were statistically significantly associated with any of the specific clinical signs or outcomes investigated (p-values>0.05) except death among NL63 cases (Table 2); however, we had limited power to detect associations given the small number of HCoV positive cases.

Co-infection with other respiratory viruses
About 47% (n=147) of the 314 HCoV cases were co-infected with other viral respiratory pathogens; respiratory syncytial virus (RSV) and human rhinovirus (HRV) jointly accounted for >50% of all HCoV coinfections with other pathogens. A similar coinfection pattern was observed for each HCoV tested ( Figure 2). Throughout the surveillance period, there were three cases (one of NL63 and two of OC43) aged <1 year that were readmitted and tested positive for the same HCoV as the first admission. The NL63 readmission occurred 10 days after discharge from the first admission while the OC43 readmissions were at 3 and 21 days after discharge from the first admission. The NL63 case had a discharge diagnosis of neonatal sepsis for the first admission and gastroenteritis plus lower respiratory tract infection (LRTI) for the readmission. One of the OC43 cases had a discharge diagnosis of LRTI for both admissions while the other had immunosuppression plus malnutrition in the first and immunosuppression plus septicaemia for the second admission.
Temporal patterns of different HCoVs NL63 and OC43 were observed fairly consistently throughout the surveillance period while fewer cases of HCoV-229E were observed from the middle of 2011 and were not detected subsequent to 2016 ( Figure 3). The highest numbers of cases were observed in the periods April to June for NL63, June to September for OC43 and January to March for 229E. Pooling data for all HCoVs, there were more cases in the colder months (May to September) than the hotter months (October to April) ( Figure 4), as for OC43, but NL63 was more common in the first half of the year, and 229E in the second half of the year. However, time series models did not indicate a seasonal pattern for any of the HCoVs ( Figure 5) over the years. The proportion of samples tested for HCoV did not change over time among those with severe pneumonia ( 2 (1) 3.11; χ = p-value = 0.078) but changed among those with very severe pneumonia 2 (1) 149.11; χ = p-value < 0.001).
De-identified raw data for this study are available as Underlying data 18 .

Discussion
We have described the circulation patterns of three endemic HCoVs (NL63, OC43 and 229E) in a long-term surveillance study of childhood pneumonia hospitalisations in coastal Kenya. We observed a small proportion of pneumonia admissions positive for one or more HCoVs (3.9%). While 65% of HCoV infections occurred in children in their first year of life (either cumulatively for all HCoVs or for each individual HCoV type), this reflected the age-distribution of pneumonia admissions to the ward. Hence, contrary to other reports 19 , HCoV did not differ by age and this might reflect no variation in age-specific community incidence of infection in children under five, together  with lack of variation in disease severity (likelihood of hospitalisation) across ages, or (ii) variation in age-specific community incidence by age but with a disproportionate probability of hospitalisation. Our reported prevalence is equivalent to that from a long-term hospital surveillance of seasonal coronaviruses in Scotland (4%) 8 7 . This latter study included locations with a wide range of climate conditions that might influence prevalence; however, the study was not big enough to stratify by location.
We did not observe seasonal variation of HCoVs compared to some other respiratory viral pathogens like RSV, as previously reported from our site 14 . In addition, neither peak months for pneumonia admissions nor the long rainy periods (April to July) in Kilifi were associated with HCoV peaks. This is in contrast to data from temperate settings where seasonality of HCoVs has been reported 8,[20][21][22] , with increased occurrence during the colder winter months. The HCoVs we have studied are known to continuously circulate among humans 1 , although in Kilifi we observed low numbers for all types. Of interest is that pneumonia associated with 229E admissions was not detected in the later years of the surveillance. We attempted to investigate if this was due to primers or probe mismatches. For all the three tested endemic CoVs we did not observe significant mismatches on the primer/probe pairs against data available from GenBank database although this investigation suffered a limitation of few sequences available globally in recent years (2015-2019) and none from East Africa. With the highest numbers and consistent presence compared to NL63 and 229E, our results suggest that OC43 is the predominant HCoV type in the coastal region of Kenya.
The present study did not have a control group by which to assess an aetiological association between the HCoVs and pneumonia. In the PERCH multi-country case-control study 17 HCoVs contributed less than 1% of the etiological fraction. In our study the contribution to disease is not known (except for the relatively small set of samples from 2011-13 that were part of the PERCH study), but it is of note that around 50% co-occurred with another respiratory virus (most commonly with  RSV), the risk was not age-dependent, there was no clear association between any of the viruses with the any of the specific clinical signs or outcomes investigated and in 26% of deaths with a HCoV detected there was a likely alternative diagnosis to pneumonia. This suggests the presence of incidental HCoV in Kilifi and other clinical groups (hence an under-estimation in this study), though causality of death outcomes was not differentially identified in this study. While we have been able to sequence the virus from a proportion of the positive specimens 23,24 , we cannot assume 100% specificity, and even a modest level of false positivity could account for many of the positive diagnoses and argues for caution in interpreting the prevalence estimates. Of relevance also is that few (~1%) of the 314 children positive for at least one HCoV were subsequently HCoV-positive readmissions. While this is a crude analysis which ignores censoring at the start and end of the surveillance, and alternative hospitals where patients may have been admitted, it might be an indicator of low probability of severe reinfection. Alternatively, these readmissions might be children with a prolonged infection having not fully recovered at their first discharge. In addition, HKU1 was only tested during the 24 month period that PERCH study was conducted hence   samples tested for HKU1 were considered unrepresentative of the entire study period and therefore not included in this analysis.
Over the surveillance period, we have changed our sample collection and testing methods. This is a limitation; we did not conduct a sensitivity analysis to compare the different PCR methods, and the addition of an OP swab increases the number of viruses found by NP alone (by 14% for HCoVs) 15 . Some of the observed patterns may have been influenced by these changes. It should be noted that fever, either on history or as measured at the time of admission, was not an inclusion criterion for eligibility, which might have influenced the prevalence of HCoV. If the endemic COVs are strongly associated with symptoms of fever, as is SARS CoV-2, then we might be under-estimating the prevalence due to a selection bias. However, interestingly, only 58% of the HCoV positive cases had axillary temperature of >=37.5°C. A further limitation is that only a fraction (70%) of pneumonia cases was tested for HCoV. We have previously shown that those untested tend to be more severely ill and less likely to be virus positive 14 . The proportion tested has not substantially changed over time among individuals with severe pneumonia but changed among those with very severe pneumonia. Similarly, we did not observe a difference in the ages of those tested and those untested for HCoV across time. We employed various collection methods and these are known to differ in the range of viruses detected. For example, the addition of an OP swab to an NP swab has been found to increase virus yield 15 .
In conclusion, in this tropical setting we find the three endemic coronaviruses were observe at low prevalence, not dissimilar to influenza and metapneumovirus, but considerably lower that for RSV and human rhinovirus. There was no clear seasonal variation. As the pandemic of COVID-19 takes its course, it is of interest to speculate whether the SARS-CoV-2 virus will become endemic and continuously co-circulate in the human population with the existing HCoVs 7,25 . The epidemiology of endemic HCoVs can be used to inform our expectations of SARS-CoV-2 in childhood, its potential severity and inter-species interactions and competition.

Open Peer Review
I am satisfied that the authors have sufficiently addressed all the concerns raised.
One minor comment: reference #8 is available in peer reviewed published form if the authors wish to update their reference list: https://doi.org/10.1093/infdis/jiaa185 1 .
(since this can affect probabilty of virus detection and violates statistical assumptions of independent data points).
Suggest authors add to Methods a brief outline of the changes in testing procedure over time (as mentioned in the Discussion) to aid interpretation of detection trends, as different specimens are likely associated with different probabilities of virus detection.

2.
It is unclear why the laboratory screen omitted HCoV-HKU1. Suggest the authors include a comment on the anticipated influence on the overall detection of HCoV. 3.
It is unclear whether exclusion of untested discharged patients (presumably milder cases) have led to an under-estimation of HCoV detection proportions. Suggest authors add a statement to describe any anticipated influence of this exclusion.

4.
The % of pneumonia cases with HCoV detected is clearly low. However, it would be helpful if the authors could discuss this result in the context of other common respiratory viruses detected among cases of pneumonia in this setting, such as influenza and RSV.

5.
Discussion page 6: "...suggests age is not a risk factor for coronavirus associated pneumonia hospital admission". The role of age as a risk factor for pneumonia cannot be determined here without a non-pneumonia control group. Suggest authors reword this statement and consider whether the lack of variation in age-specific proportions may potentially reflect (i) no variation in age-specific community incidence of infection in children under five, together with lack of variation in severity (likelihood of hospitalisation) across ages, or (ii) variation in age-specific community incidence by age but with a disproportionate probability of hospitalisation. 6.
Discussion page 9: "...and in 26% of deaths with a HCoV detected there was a likely alternative diagnosis to pneumonia". This seems to suggest the role of HCoV on overall hospital burden may be under-estimated by pneumonia cases. Although not an aim of the study, I suggest the authors consider adding a brief comment regarding the generalisability of their findings to other clinical groups to place the results into a wider context.

7.
Discussion page 9: Can the authors comment on the possibility that readmissions reflect prolonged virus shedding, rather than indicating reinfection? Longest duration between admissions of 21 days possibly bordering an anticipated peak detection within first few weeks (e.g. for SARS-CoV-2 1 ). Suggest authors consider adding a brief statement to include this possibility. 8.
Discussion page 10: " fever... was not an inclusion criterion for eligibility, which might have influenced the prevalence of HCoV". Suggest authors make clear the anticipated direction of effect on the prevalence -presumably underestimation? 9.
Discussion page 10: "...there was a significant age difference for those tested and those untested for HCoV across time." This statement seems to be at odds with Results page 3 "..cases untested did not differ from those tested in age distribution.." Suggest the authors reword as appropriate. 10.
Conclusions page 10: "...little evidence of a substantial aetiological contribution...": The aetiological role of HCoV cannot be established without a control group (as the authors do mention on page 9) i.e. it is not known whether the prevalence of HCoV is lower among cases of ARI without pneumonia. Suggest authors rephrase to put the low HCoV detections within the context of relative contribution of pathogens/other causes to disease aetiology i.e. most pneumonia cases are attributable to other causes.

11.
Although presumably outside the scope of this study to analyse, I suggest the authors consider commenting briefly on any potential role of coinfecting respiratory bacteria on HCoV-associated pneumonia in this population; bacterial coinfection may influence the chance of pneumonia/testing for viruses. 12.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Partly Discussion page 6: "...suggests age is not a risk factor for coronavirus associated pneumonia 1.
hospital admission". The role of age as a risk factor for pneumonia cannot be determined here without a non-pneumonia control group. Suggest authors reword this statement and consider whether the lack of variation in age-specific proportions may potentially reflect (i) no variation in age-specific community incidence of infection in children under five, together with lack of variation in severity (likelihood of hospitalisation) across ages, or (ii) variation in age-specific community incidence by age but with a disproportionate probability of hospitalisation.

Response:
We have made the change as suggested by the reviewer.
Discussion page 9: "...and in 26% of deaths with a HCoV detected there was a likely alternative diagnosis to pneumonia". This seems to suggest the role of HCoV on overall hospital burden may be under-estimated by pneumonia cases. Although not an aim of the study, I suggest the authors consider adding a brief comment regarding the generalisability of their findings to other clinical groups to place the results into a wider context.

Response:
We have made the change as suggested by the reviewer.
Discussion page 9: Can the authors comment on the possibility that readmissions reflect 1.
prolonged virus shedding, rather than indicating reinfection? Longest duration between admissions of 21 days possibly bordering an anticipated peak detection within first few weeks (e.g. for SARS-CoV-21). Suggest authors consider adding a brief statement to include this possibility.
Response: It is possible these are prolonged infections rather than reinfections and we have added the following to the text in the discussion: "...probability of severe reinfection. Alternatively, these readmissions might be children with a prolonged infection having not fully recovered at their first discharge" Discussion page 10: " fever... was not an inclusion criterion for eligibility, which might have 1.
influenced the prevalence of HCoV". Suggest authors make clear the anticipated direction of effect on the prevalence -presumably underestimation?
Response: This is possible, but we don't know that these endemic HCoVs behave like SRAS-CoV-2 ie are strongly associated with fever, hence again difficult to speculate as to which direction. We therefore added this statement to the limitations " If the endemic COVs are strongly associated with symptoms of fever, as is SARS=CoV-2, then we might be underestimating the prevalence due to a selection bias".
Discussion page 10: "...there was a significant age difference for those tested and those untested for HCoV across time." This statement seems to be at odds with Results page 3 "..cases untested did not differ from those tested in age distribution.." Suggest the authors reword as appropriate.

Response:
We have made the change as suggested by the reviewer.
Conclusions page 10: "...little evidence of a substantial aetiological contribution...": The aetiological role of HCoV cannot be established without a control group (as the authors do mention on page 9) i.e. it is not known whether the prevalence of HCoV is lower among cases of ARI without pneumonia. Suggest authors rephrase to put the low HCoV detections within the context of relative contribution of pathogens/other causes to disease aetiology i.e. most pneumonia cases are attributable to other causes.

Response:
We have made the change as suggested by the reviewer.
Although presumably outside the scope of this study to analyse, I suggest the authors consider commenting briefly on any potential role of coinfecting respiratory bacteria on HCoV-associated pneumonia in this population; bacterial coinfection may influence the chance of pneumonia/testing for viruses.
Right column, second line: "… alive or dead)." ○ There are two mentions of which viruses were tested for on page 3: suggest concatenate into one mention (probably best placed in the "Laboratory methods" paragraph). Reword at the bottom of page 3 on the right hand column to "… or very severe), laboratory test results, and hospitalisation outcomes…" 10.
The link to code in Extended data is not currently active in the manuscript: please correct this.

11.
At the bottom of page 3, suggest reword as "Untested cases did not differ… more likely to have very severe pneumonia (40.0% versus 28.3%; Χ 2 =152.5, P<0.001)." Why resort to the use of Fisher's exact P-value in the analysis of tested and untested cases (the Chi-square P-value is every bit as "significant")?
for (3.9%)." Check the calculation of the comparison between Sipulwa et al and the current study, in the second line on the right hand column on page 9: I get Χ 2 =3.92, p=0.048.

24.
In the second sentence of the second paragraph on the right column on page 9, suggest reword to "… nor the long rainy periods (April to July) in Kilifi were associated with HCoV peaks."

25.
Suggest reword the fifth sentence in the right hand column on page 9 as "Of interest is that pneumonia associated with 229E was not detected in the later years of surveillance."

26.
Please correct the spelling or "axillary" in the footnote to Table 2. 27.
The authors do not consider differences in participant recruitment as being a potential explanation as to why 229E-associated pneumonia was not detected subsequent to 2016: were there systematic differences in participant recruitment strategies that might have impacted on detection of this virus?

28.
Suggest reword the third sentence on the right hand column on page 10 as "It should be noted that fever, either on history or as measured at the time of admission, was not an inclusion criterion…" 29.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Partly