The interactions of SARS-CoV-2 with cocirculating pathogens: Epidemiological implications and current knowledge gaps

Despite the availability of effective vaccines, the persistence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) suggests that cocirculation with other pathogens and resulting multiepidemics (of, for example, COVID-19 and influenza) may become increasingly frequent. To better forecast and control the risk of such multiepidemics, it is essential to elucidate the potential interactions of SARS-CoV-2 with other pathogens; these interactions, however, remain poorly defined. Here, we aimed to review the current body of evidence about SARS-CoV-2 interactions. Our review is structured in four parts. To study pathogen interactions in a systematic and comprehensive way, we first developed a general framework to capture their major components: sign (either negative for antagonistic interactions or positive for synergistic interactions), strength (i.e., magnitude of the interaction), symmetry (describing whether the interaction depends on the order of infection of interacting pathogens), duration (describing whether the interaction is short-lived or long-lived), and mechanism (e.g., whether interaction modifies susceptibility to infection, transmissibility of infection, or severity of disease). Second, we reviewed the experimental evidence from animal models about SARS-CoV-2 interactions. Of the 14 studies identified, 11 focused on the outcomes of coinfection with nonattenuated influenza A viruses (IAVs), and 3 with other pathogens. The 11 studies on IAV used different designs and animal models (ferrets, hamsters, and mice) but generally demonstrated that coinfection increased disease severity compared with either monoinfection. By contrast, the effect of coinfection on the viral load of either virus was variable and inconsistent across studies. Third, we reviewed the epidemiological evidence about SARS-CoV-2 interactions in human populations. Although numerous studies were identified, only a few were specifically designed to infer interaction, and many were prone to multiple biases, including confounding. Nevertheless, their results suggested that influenza and pneumococcal conjugate vaccinations were associated with a reduced risk of SARS-CoV-2 infection. Finally, fourth, we formulated simple transmission models of SARS-CoV-2 cocirculation with an epidemic viral pathogen or an endemic bacterial pathogen, showing how they can naturally incorporate the proposed framework. More generally, we argue that such models, when designed with an integrative and multidisciplinary perspective, will be invaluable tools to resolve the substantial uncertainties that remain about SARS-CoV-2 interactions.


Introduction
As of August 2022, the pandemic of coronavirus disease 2019 (COVID-19)-caused by the novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)-has resulted in at least 598 million cases and 6.4 million deaths worldwide [1]. Despite the implementation of stringent control measures and the increasing roll-out of effective vaccines in many locations, the persistent circulation of SARS-CoV-2 suggests the infeasibility of elimination and the gradual transition to endemic or seasonal epidemic dynamics [2]. Hence, co-circulation of SARS-CoV-2 with other pathogens may become increasingly frequent and cause multiple simultaneous epidemics of, for example, COVID-19 and influenza [3].
Interaction-that is, the ability of one pathogen to alter the risk of infection or disease caused by another pathogen (Fig. 1)-is an essential aspect to forecast the dynamics of cocirculating infectious diseases. From a public health perspective, interactions may significantly aggravate disease burden, as demonstrated for immunosuppressive viruses like measles [4] and human immunodeficiency virus (HIV) [5]. Another interesting, yet understudied public health implication of interactions is the possibility of indirect effects of vaccines on non-target pathogens, as suggested for influenza vaccines [6,7] . However, despite their potentially large relevance to SARS-CoV-2 epidemiology and COVID-19 control measures, the interactions of SARS-CoV-2 with other pathogens remain poorly defined.
Here, we aimed to review the current body of evidence about the interactions of SARS-CoV-2 with co-circulating pathogens. We first present a general framework to capture the complexities of interactions and study them in a systematic way. Using this framework, we then review the results of published experimental and epidemiological studies. Finally, we formulate simple transmission models incorporating the proposed framework to illustrate the potential population-level impact of SARS-CoV-2 interactions. 5

Dissecting pathogen interactions: Sign and Strength, Timing, and mechanisms
Pathogen interactions can be complex, because of the multiple elements needed to fully characterize them. To study interactions in a systematic and comprehensive way, we propose a conceptual framework-depicted schematically in Fig. 1-that incorporates three essential components of interaction, detailed below.

Sign and strength of interaction
The first dimension of this framework is the sign and strength of interaction. Here, we define the sign of interaction as positive in synergistic interactions (where a first pathogen increases the risk of infection or disease of a second pathogen) and negative in antagonistic interactions (where the risk is decreased), and we refer to strength as the magnitude of effect on a given parameter exerted by one pathogen on another.
An example of negative interaction exists between influenza A virus (IAV) and human respiratory syncytial virus (RSV), for which experimental studies have shown that a recent IAV infection inhibits the growth of RSV in ferrets [8] and in mice [9]. By contrast, IAV interacts positively with Streptococcus pneumoniae (the pneumococcus) by promoting bacterial growth [10,11]. This illustrates that interaction is pathogen-specific and cannot be easily extrapolated to other pathogen systems. 7 The second dimension of our proposed framework is time-dependency: both the time between infections and the sequence of infection can affect the sign and strength of an interaction.

Duration of interaction and time between infections
Due to the kinetics of cellular and humoral immune response following respiratory infections [12][13][14], the strength of interaction can change with time between infections. For example, primary IAV infection prevented subsequent RSV infection in ferrets when exposed 3 days later but the protection disappeared as the time between IAV and RSV challenges increased to 11 days [8]. Such short-lived negative interaction was also observed between influenza B virus Victoria lineage (B/Vic) and Yamagata lineage (B/Yam) [15]. Interaction can be long-lived if it is mediated by immune memory. For example, measles infection can partially erase previously acquired immunity to other pathogens, causing "immune amnesia" [16].
Childhood exposures to a given IAV subtype can cause long-lasting immunological bias that shapes the individual's subsequent risk for influenza infection [17].

Symmetry of interaction and sequence of infection
The sequence of infection can also affect the interaction, as evidenced by the asymmetric effects found in previous studies. For example, prior infection with IAV or RSV hindered rhinovirus (RV) replication, but prior RV infection did not interfere with IAV and RSV replication in human airway epithelium [18]. While IAV infection predisposed individuals to pneumococcal colonization and infection [19][20][21] and led to more severe disease [22], evidence from animal and human challenge studies demonstrated that prior pneumococcal colonization did not lead to more severe disease [20,23,24] but might have had a protective 8 effect against viral replication [24,25] upon subsequent IAV challenge. Interestingly, this effect might depend on the density of pneumococcal colonization [20,23,24].
By contrast, when a negative interaction is symmetric between two pathogens, whichever of the two pathogens is the first to infect can inhibit subsequent infection by the other pathogen-as in the case of influenza B lineages [15].

Biological mechanisms and population-level impact of interaction
The third dimension in our framework is the mechanism of interaction: interaction can be caused by different biological mechanisms, which determine its positive or negative effects on susceptibility to infection, characteristics of infection (such as transmissibility and duration), or disease severity at the individual level and in turn its impact at the population level (Fig.   1C).

Biological mechanisms
Examples of biological mechanisms of pathogen interaction include intra-cellular and physiological changes and effects on the immune response, on the respiratory microbiota, and on host behaviors. A pathogen can induce changes on the host cells that are beneficial or detrimental to another pathogen. For example, it has been shown that RSV and human parainfluenza virus 3 (HPIV-3) increase expression of receptors for Haemophilus influenzae and the pneumococcus binding in bronchial epithelial cells [26]. In both cases, changes in cellular expression may lead to a positive interaction. A pathogen can cause changes to the host's immune profile (e.g., depletion of CD4+ T cells by HIV [5], increased IFN response by IAV [9]), facilitating or hindering infection with a second pathogen. Moreover, a pathogen can change the physiological environment to potentiate a secondary infection by another pathogen. For instance, the replication of IAV in the respiratory epithelium reduces 9 mucociliary clearance and damages epithelial cells, resulting in enhanced attachment and invasion of the pneumococcus [21]. Changes in the respiratory tract microbiota by an infection can lead to the acquisition of a new pathogen, or to overgrowth and invasion of an already present pathogen [27][28][29]. Lastly, changes in host behaviors caused by infection with a first pathogen can affect the risk of subsequent infection with another pathogen, even in the absence of within-host interaction between the two. Examples include self-isolation to reduce spread of disease in humans and reduced social contacts in infected animals [30,31].

Population-level impact
The biological mechanisms outlined above may affect population-level dynamics through their effects on different epidemiological parameters: susceptibility to infection, transmission of infection (characterized by the transmissibility and the duration of infection), and disease severity. Of note, multiple biological mechanisms can affect the same epidemiological parameter; conversely, the same biological mechanism can affect multiple epidemiological parameters. For example, IAV-induced epithelial damage and dampened pneumococcal clearance increase host susceptibility to the pneumococcus and disease severity in co-infection, as suggested by historical pandemics [32], demonstrated in experimental studies [19], and inferred from mechanistic modeling of epidemiological time-series [33,34].
The effect of interaction on transmission is more difficult to measure, as it is determined not only by the susceptibility of the exposed and the transmissibility of the infected, but also by the   ; all the corresponding values were checked and are available in Tables   S1 and S2. Having proposed a framework to study interactions, we now review experimental studies on co-infections with SARS-CoV-2 in animal models. As of August 22th, 2022, we identified 14 publications [43-54,56-58]. We first review the 11 studies that focused on SARS-Cov_2 and non-attenuated IAV.
Experimental studies of co-infection with SARS-CoV-2 and non-attenuated IAV As shown in Fig. 2, three different animal models were used (ferrets, hamsters, and mice) and the experimental designs varied substantially across the eleven studies, particularly in the sequence of infection, the time between infections (range: 0-21 days), and the follow-up  12 As shown in Fig. 3A, the severity of mono-infection with either IAV or SARS-CoV-2 differed between animal models. In ferrets, mono-infection with IAV, but not with SARS-CoV-2, resulted in weight loss, while the opposite was observed in hamsters. In mice, however, both mono-infections generally caused weight loss. Also unlike the hamster and ferret models, mice can develop severe COVID-19 and die, so that this model was used in all studies that analyzed survival (Fig. 3B). On the whole, these results agree with earlier evidence of the advantages and limitations of different animal models for in vivo research on IAV and SARS-CoV-2 [59,60].
In all but one study, the effect of co-infection on disease severity was quantified by tracking changes in the animals' body weight. In mice and, to a lesser extent, in hamsters, animals co-infected suffered a higher maximal weight loss than animals mono-infected with either IAV or SARS-CoV-2 (Fig. 3A, Table S1). In ferrets, however, the maximum weight loss after co-infection was relatively comparable to that after IAV mono-infection. In keeping with the results based on weight loss, the three studies that measured survival (all using the mice model) found that co-infected animals either suffered higher mortality [45,50] or died faster [44] than mono-infected animals ( Fig 3B, Table S1).
In contrast to the relatively consistent results on disease severity, the effect of coinfection on viral load-quantified as the ratio of viral load during co-infection to that during mono-infection-was more heterogeneous across studies (Fig 4, Table S2). In addition to the sources of heterogeneity outlined above, the studies varied in the technique used to quantify viral load (either RT-qPCR, plaque-based or median tissue culture infectious dose (TCID50) assays) and in the sample type (swabs or tissue) and location (lower respiratory tract (LRT) or upper respiratory tract (URT)). These differences may affect the inferred sign and strength of interaction: for example, the load of infectious viruses-which only plaque-based or TCID50 assays can quantify-in the URT is likely a more relevant proxy of transmissibility [61], but 13 was measured in only six studies [45,46,49,[51][52][53]. Overall, the effect size spanned six orders of magnitude and depended on the location of the body compartment sampled. In the LRT, the viral load of SARS-CoV-2 was generally reduced by preceding or simultaneous infection with IAV, but increased by subsequent infection with IAV in hamsters (Fig. 4A, left panel). The effect was more variable in mice, and inconclusive in ferrets because of a low number of studies. On the other hand, there was no obvious pattern in the viral load of IAV, regardless of infection order (Fig. 4A, Table S1) and the effect on viral load in one study ([51], Fig. 4A, Table S2).
In conclusion, despite large heterogeneity and inconsistencies across the studies reviewed, the collective evidence from animal models shows that co-infection with IAV and SARS-CoV-2 causes more severe disease than mono-infection with either virus. Despite having clinical relevance, these results do not necessarily demonstrate a positive interaction. This is because the endpoints in all studies were non-specific, making it difficult to hypothesize the expected disease severity resulting from mere co-occurrence of two independent infections that do not interact. Virus-specific endpoints are therefore needed to conclusively demonstrate an interaction affecting disease severity. Despite the availability of such endpoints to assess the effect of co-infection on viral load, the collective evidence was inconclusive. A generally robust finding was that preceding or simultaneous infection with IAV reduced the viral load of SARS-CoV-2 in the LRT. However, only a few studies measured the viral load in the URT, which is likely a more relevant proxy of transmissibility [61]. Therefore, further studies will 14 be needed to demonstrate the existence of interactions affecting susceptibility to, or transmissibility of, infection. In the design of such studies, we argue that the strength of evidence could be increased by varying the infectious dose and the infection order, and by considering different animal models.   all the corresponding values were checked and are available in Tables S1 and S2. 17

Epidemiological evidence
Although experimental studies using animal models can inform some of the components required to characterize pathogen interactions (Fig. 1 (Table S3). All three studies involving PPSV did not find conclusive evidence for association between PPSV history and SARS-CoV-2 related outcomes [88,89]. PCV was associated with protection against COVID-19 infection, hospitalization, and mortality among older adults in one cohort study [89], and against symptoms among SARS-CoV-2-infected children in another cohort study [91]. Although inconclusive, the association estimated in a case-control study [88] was consistent with that in the two cohort studies.  1.55-9.21 for 1-14 days prior, OR: 3.59, 95% CI: 1.42-9.05 for 1-30 days prior) in a casecontrol study [99]. This evidence, suggestive of a positive interaction between influenza and SARS-CoV-2, is consistent with the findings from a mathematical modeling study [103].
Although a retrospective cohort study reported that prior infection with endemic human coronaviruses (hCoVs) was associated with protection against COVID-related ICU admission . This discrepancy may be explained by the different URI definitions and time frames for exposure measurement, in addition to different study designs and included confounders. Because these studies provided information about the infection timeline, they offered stronger evidence to infer pathogen interactions than studies based on co-infection prevalence, and also more direct evidence than studies examining the association between non-COVID vaccines and COVID-19. Nevertheless, one should beware of how misclassification of exposure and imperfect control for confounding can limit such study designs in inferring pathogen interactions.
In summary, the evidence available from human population health data indicates that co-infection prevalence is largely variable, that influenza vaccines and PCVs may be associated with reduced risk of SARS-CoV-2, and that earlier influenza infection may be associated with higher risk of SARS-CoV-2 infection and disease severity. However, our review also highlighted the limitations in the current epidemiological literature, as many studies were prone to multiple biases, including confounding, and only very few [99-103] were designed to infer interaction.

The need for transmission models to study epidemiological interactions
As reviewed above, the results of epidemiological studies can be difficult to interpret and their designs insufficient to characterize all the components of interactions ( Fig. 1).
Arguably, more integrated approaches are therefore needed to capture the complexities described above and to determine how individual-level mechanisms of interaction translate into population-level dynamics of infection or disease.
Mathematical models of transmission offer a powerful and economical tool to study infectious disease dynamics [104]. To study pathogen interactions, such models can be formulated to incorporate biologically explicit mechanisms of interaction (in addition to the other elements of the framework proposed above) and predict their potentially non-linear 22 effects on transmission dynamics [105]. By design, these models translate between scales, such that the population-level impact of a given individual-level mechanism of interaction can be simulated and predicted. To illustrate the relevance of such models, we formulated two basic models of SARS-CoV-2 interaction (see more details and equations in the Supplement), with either an endemic colonizing bacterium (e.g., the pneumococcus) or a respiratory virus causing seasonal epidemics (e.g., influenza). In both cases, we assumed a non-symmetric (i.e., no effect of SARS-CoV-2 on the other pathogen) interaction that caused a 1-5 fold (strength) decrease or increase (sign) of SARS-CoV-2 transmission (mechanism) from co-infected individuals (duration of interaction equal to the infectious period of the other pathogen). Importantly, the within-host processes causing interaction were not explicitly modeled, but their effects were represented by these interaction parameters. As shown in Fig. 5A, we find that even a moderately strong interaction with a commensal bacterium can substantially affect the dynamics of SARS-CoV-2, increasing its peak incidence by 3.5 fold for positive interaction when the prevalence of bacterial colonization reaches 50% of the population (as frequently observed in young children for the pneumococcus[106,107]). By contrast, an equal interaction with an epidemic virus is predicted to have a much smaller impact on the dynamics of SARS-CoV-2 ( Fig. 5B). Of note, the maximal impact is predicted at intermediate levels of transmissibility of the epidemic virus, corresponding to maximal epidemic overlap with SARS-CoV-2 (Fig. 5B). This finding emphasizes a major difference between endemic and epidemic pathogens: for the latter, the impact of even strong interactions may remain subtle and manifest itself only after a prolonged period of co-circulation with SARS-CoV-2. Overall, these numerical experiments demonstrate the value of mathematical models to study interactions in a biologically explicit and comprehensive way and to predict their complex (and potentially unexpected) effects at the population level.

23
Although voluntarily over-simplified and used here only for illustrative and exploratory purposes, these models can be readily extended to add components relevant to SARS-CoV-2 epidemiology, such as age, vaccination, or temporal variations in transmission caused by new variants, seasonality, or changing control measures. In real-world applications, however, model parametrization can be a substantial challenge, as the values of many parameters may be neither directly observable nor fixed from empirical evidence. This problem is particularly salient for parameters characterizing interaction, whose values can be only partially inferred from

Conclusion
As population immunity against COVID-19 accrues in many regions worldwide, it is critical to understand the factors that will affect the future transmission dynamics of SARS-CoV-2 [2]. Here, we proposed that interactions with co-circulating pathogens will be such a key factor. Indeed, such interactions may have notable public health implications, in particular for forecasting and controlling SARS-CoV-2 epidemics and for predicting the indirect impact of vaccines. The scientific implications of interaction are also notable and may lead to considering SARS-CoV-2 as part of polymicrobial systems whose individual components cannot be well studied separately.
Despite the relevance of interaction, our review identified only a few experimental studies in animal models, with markedly different designs and the majority focusing on SARS-CoV-2 interaction with IAV. A robust finding from our comparative analysis is that SARS-

List of supplementary materials
Appendix 1: Table S1. An overview of the experimental designs and results on disease severity, measured as maximal body mass loss or survival at experiment end, from the reviewed studies assessing the interaction between SARS-CoV-2 and influenza A virus (IAV).
Appendix 2: Table S2. An overview of the experimental designs and results on viral load, measured in the upper or lower respiratory tract, from the reviewed studies assessing the interaction between SARS-CoV-2 and influenza A virus (IAV).
Appendix 3: Table S3. Observational studies examining the association between pneumococcal vaccination history and COVID-19.
Appendix 4: Table S4. Observational studies examining the association between prior respiratory infections and COVID-19.
Appendix 5: Model details (including Figure S1: Schematic of the bacteria-virus interaction model; Table S5. Parameters used for S. pneumoniae -SARS-CoV-2 interaction model; Figure  S2: Schematic of the virus-virus interactions model; and Table S6. Parameters used for influenza A -SARS-CoV-2 interaction model).

Table of Contents
Appendix 1: Table S1. An overview of the experimental designs and results on disease severity, measured as maximal body mass loss or survival at experiment end, from the reviewed studies assessing the interaction between SARS-CoV-2 and influenza A virus (IAV). Table S2. An overview of the experimental designs and results on viral load, measured in the upper or lower respiratory tract, from the reviewed studies assessing the interaction between SARS-CoV-2 and influenza A virus (IAV). Table S3. Observational studies examining the association between pneumococcal vaccination history and COVID-19. Table S4. Observational studies examining the association between prior respiratory infections and COVID-19.

Appendix 5:
Model details (including Figure S1: Schematic of the bacteria-virus interaction model; Table S5. Parameters used for S. pneumoniae -SARS-CoV-2 interaction model; Figure   S2: Schematic of the virus-virus interactions model; and Table S6. Parameters used for influenza A -SARS-CoV-2 interaction model).

Max body mass loss
Appendix 1: Table S1. An overview of the experimental designs and results on disease severity, measured as maximal body mass loss or survival at experiment end, from the reviewed studies assessing the interaction between SARS-CoV-2 and influenza A virus (IAV). Values were taken from tables or text, or when these were not available, extracted from the figures using the program PlotDigitizer [12].  1 and Z the state of pathogen 2 [1].
Bacteria-virus interaction model: The bacteria-virus model was constructed such that pathogen 1 is the bacteria and pathogen 2 is the virus. We assumed the interaction was asymmetric, such that colonization with bacteria impacts transmission of the virus, but infection with the virus has no impact on the bacterial dynamics. The model was defined by 2 x 3 = 6 ordinary differential equations as represented in Figure S1, where the disease states are {S, C} for bacteria, and {S, I, R} for the virus. As an example we consider the S. pneumoniae -SARS-CoV-2 interacting system, hence assuming the bacteria is S. pneumoniae and the virus is SARS-CoV-2. Parameter values are detailed in Table S5. The model was run for 365 days and the peak viral incidence was calculated for varying rate of bacterial colonization and varying transmission interaction parameter. The model was implemented in the R [2] packages 'pomp' [3], and 'tidyverse' [4].
All code is available at https://github.com/egoult/pathogen_coinfections .   Here, / 0,1 and / 0,2 denote the respective basic reproductive numbers for virus 1 and virus 2. We consider the Influenza A -SARS-CoV-2 interacting system as an example, where virus 1 is Influenza A and virus 2 is SARS-CoV-2, so infection with influenza A affects the dynamics of SARS-CoV-2, but infection with SARS-CoV-2 has no impact on influenza A. Parameter values are detailed in Table S6. The model was run for 365 days and the peak SARS-Cov-2 incidence was calculated, for varying of influenza A basic reproduction numbers and varying transmission interaction parameter.