3.Results
3.1. Spontaneous cytokine release signatures in PBMCs reflect respiratory virome profiles.
Based on the baseline release of 13 out of 22 measured cytokines, individuals were annotated in principal components (Figure 1 and Supplementary Data 2). Unsupervised clustering identified two groups with distinct cytokine release signatures (Figure 1). The first group (Cluster 1) included individuals (n=37, 72.5%) that were characterized by low spontaneous cytokine release (Figure 1a-1b), while the second group (Cluster 2) included individuals (n=14, 27.5%) with high overall spontaneous cytokine release (Figure 1a-1b). All individual cytokines contributed to the clustering at a range of 5.5%-7.2% (Supplementary Figure 1a). The majority (86%) of children in the high cytokine spontaneous release group (Cluster 2) had prokaryotic dominated viromes (PVPG)(Specifically, high Shannon diversity and richness of prokaryotic viruses), in contrast to the low cytokine spontaneous release group (cluster 1) (Figure 1c) that included evenly distributed children with all three types of virome (PVPG, EVPG, and AVPG) (p:0.018, Chi-Square test). These findings were independent of the presence of allergy or asthma (Supplementary figure 1b-1c). There were no significant associations between baseline cytokine clusters and the presence of specific viral families (Picornaviridae, Anelloviridae, and Siphoviridae) in the upper respiratory tract virome (Supplementary Figure 2). To further assess baseline responses, a logistic regression model considering geography (Greece, Poland and Finland), VPG, sex, age and presence of Picornaviridae and Anelloviridae was performed (Supplementary Table 3). This showed that EVPG and AVPG were statistically significant and negatively associated with PVPG (EVPG p:0.034, beta: -2.92; AVPG p:0.022, -3.01) in the non-stimulated culture medium (Supplementary Table 4), confirming that the virome groups reflected the two types of response.
3.2. PBMC bacterial immune signatures have minimal associations with respiratory virome characteristics.
Bacterially-stimulated samples were annotated in a coordinate plot based on their cytokine induction profiles (n=17) (Figure 2a & Supplementary Data 3). Hierarchical clustering identified three clusters of antibacterial immune responses (Figure 2a and Supplementary Figure 3a). Cluster 1 (n=15, 30%) included children with low release of bacterial DNA-stimulated cytokines, despite high response to LPS, especially for IL-1b, IL-6, TNF, CCL3 and CCL4 (Figure 2b). In contrast, Cluster 2 (n=19, 37%) displayed high release of TNF, CCL3, CCL4, IL-6, IL-27 and IL-1b when stimulated by bacterial DNA (Figure 2b). Finally, Cluster 3 (n=17, 33%) displayed heterogeneous and generally low cytokine responses (Figure 2b). Each cluster was characterised according to its features: Cluster 3 children were identified as low responders, Cluster 1 as intermediate responders, and Cluster 2 as high responders. We then explored possible correlations with the presence of specific viral families (Picornaviridae (n=28, 55%), Anelloviridae (n=34, 66%), Siphoviridae (n=20, 39%)) and the virome profiles groups (VPGs) (Figure 2c) to each type of responder. Neither the presence of different virus families, nor the type of virome significantly differ across the three cytokine clusters, i.e. these were independent from the antibacterial response (Figure 2c). Moreover, antibacterial responses were not associated with asthma or allergy outcomes (Supplementary Figure 3b-3c). Nevertheless, in the regression model, the geographical location (Poland p:<0.001, beta:10.95; p:<0.001, beta:-10.91; Finland p:<0.001, beta:-9.1; p:<0.001, beta:-11.37) and the presence of Siphoviridae (p:<0.001, beta:-8.67; p:<0.001, beta:13.81) had significant inference in high and low responders respectively (Cluster 2 and 3) over intermediate responders (cluster 1) regarding their antibacterial responses (Supplementary 5), suggesting a gradient of bacteriophages in reverse correlation with responses to bacterial stimuli (i.e. high bacterial response corresponding to low levels of bacteriophages, etc).
3.3. PBMC antiviral immune signatures correlate with the presence of Siphoviridae and Picornaviridae in the upper respiratory virome.
We identified 4 clusters describing PBMC responses to viral stimuli (Figure 3a and 3b & Supplementary Data 4). IL-1b, CCL3, CCL4 and TNF in all virus-like stimulants had a major influence in clustering the samples, among a total of 23 significant conditions (Supplementary Figure 4a). Cluster 4 children (n=5, 10%) had a high and homogeneous response to TLR3 (Poly:IC) and TLR7/8 (R848) stimulated cytokines, namely IL-1b, IL-23a, IL-27, IL-6, CCL3, CCL4, TNF, IFN-a2, IL-25, TNF, IL-13, but not to rhinovirus A (RV-A) (Figure 3b). In contrast, the release of IL-1b, IFN-γ, CXCL10, CCL3, CCL4, TNF and IL-17a in response to RV-A were high in Cluster 2 children (n=11, 21.5%) (Figure 3b). Children with the lowest overall cytokine responses were grouped on Cluster 3 (n=14, n=27.5%), while the largest group of subjects displayed a heterogenous pattern in their responses and were assigned to Cluster 1 (n=21, 41%) (Figure 3b). Consequently, each group was characterized as follows: Cluster 4: overall high responders, Cluster 2: RV-A responders, Cluster 1: intermediate responses, Cluster 3: low responders.
High responders had significantly higher presence of Picornaviruses (p-value: 0.036, 95%CI) in their upper airway, in comparison to low cytokine responders (Figure 3c). This was also observed regarding Siphoviridae, however with statistically marginal value (p:0.072) (Figure 3d). No differences were observed considering the presence of Anelloviridae. The comparison of the virome profiles confirmed a significantly biased virome composition between children in the different cytokine clusters (p-value: 0.0004, 95%CI) (Figure 3c). These associations were not affected by the presence of asthma or rhinitis (Supplementary Figure 4b-4c).
In the regression model, geographical location influenced the clustering (Supplementary Table 6). The results confirmed the difference between RV-A and intermediate responders regarding to the presence of Siphoviridae and Picornaviridae (p:0.027, beta:2.83; p:<0.001, -11.63). Additionally, in low responders (Cluster 3) there were significantly more prokaryotic than eukaryotic viral group types (p:<0.001, beta:-29.67) (Supplementary Table 6).
3.4. Presence of viral families in the airway and its association with antibacterial and antiviral cytokine induction in PBMCs.
To further describe potential associations between antibacterial and antiviral PBMC responses with viral presence in the nasopharynx, we investigated cytokine induction levels in the presence of the Picornaviridae, Siphoviridae and Anelloviridae viral families (Figure 4). Among stimulants, bacterial DNA and LPS were considered for the antibacterial responses and Poly:IC, R848 and RV-A for the antiviral. When Picornaviridae were present in the nasopharynx, bacterially stimulated production of IFN-λ-2, CCL5, IL-12b were low (Figure 5). The presence of Siphoviridae was also related to low antibacterial responses (Supplementary Figure 5i), however some inflammatory cytokines (IL6, CXCL8, CCL4, TNF) were upregulated after LPS stimulation (Supplementary Figure 5ii). Regarding antiviral responses, the presence of Picornaviridae (Figure 5), as well as Siphoviridae, (Supplementary Figure 5iii) were associated with low levels of IFN-λ-2 responses; Siphoviridae were also associated with reduced IL-7, IL-23a, and IL-12b, but increased IFN-a2. In contrast, presence of Anelloviridae coincided with increased production of CCL4, IL-6, IL-27 and IL-10 against bacterial stimuli and TNF and IL-7 against viral stimuli (Supplementary Figure 6). In all, Siphoviridae displayed the broadest association with PBMC derived cytokines following either bacterial or viral stimulation (Figure 4).
4. Discussion
This is the first study showing that antiviral immune responses at a systemic level reflect the upper airway virome composition during free acute infections periods. Although most attention is currently given to the mechanisms by which the microbiota and/or their components shape the immune responses [17], we are also well aware that host immune responses can regulate microbial expansion and therefore control microbiota [18]. Analysing the immune status in combination with the microbiome is thought to be necessary for understanding the mechanisms involved in microbial influence of clinical outcome [17]. Another relevant finding of the study is the higher abundance of particular viral families, such as Picornaviridae and Siphoviridae, in individuals with low innate interferon responses.
When spontaneous cytokine release was evaluated, high producers were dominated by prokaryotic virome profiles. There are two, non-mutually exclusive, possible explanations: one, a high cytokine secretion status may result in reduction of Anellovirus and Picornavirus presence and diversity, or, high baseline responses might be the result of concurrent bacterial expansion, which in turn favours the proliferation of bacteriophages [19]. However, the latter explanation is less likely, considering that the extent PBMCs get activated by confronting bacteria during homeostasis, is minor [20].
Regarding antibacterial responses, we observed a differential response against LPS versus bacterial-DNA. Although both stimuli activate antibacterial responses in the cell, they initiate TLR signalling from distinct locations: LPS does not require internalization to activate the signal cascade, while bacterial DNA does [21]. This distinct immune activation pattern has been previously reported [22].
In line with the established understanding of rhinovirus (RV) biology, we have observed a correlation between the presence of picornaviruses (mostly RVs) and low levels of IFN-λ2, IL12 and RANTES (CCL5), following innate immune stimulation [23]. Several studies have suggested that interferon deficiency is a key mechanism supporting RV replication [24,25,26] and induction of exacerbations in patients with asthma [27]. Interestingly, although IFN-λ2 showed decreased levels in presence of picornaviruses, IFN-α2 protein levels remained unchanged. IFN-λ2 (IL-28A) belongs to IFN III-type and INF-α2 to IFN I-type. In contrast to type I IFNs, type III IFNs are not ubiquitously expressed and are mainly found at barrier epithelial surfaces such as the respiratory tract where they exhibit unique not-redundant antiviral functions [37,38].Interestingly, it has been reported that type III IFNs suppress Th2 responses in experimental asthma in mice [39], while respiratory viral pathogens have evolved mechanisms to suppress IFN-λ function or downregulate signalling, underlying their contribution to respiratory immunity at mucosal barriers [28,36].
Our findings indicate that this is a wider mechanism that controls the extent of RV presence in the upper airway mucosa, in which RV is a frequent, but transient visitor [29]. Picornaviruses were also present in all samples from subjects with high antiviral responses, but low responses to RV (cluster 4). This can be due to RV-specific defects, as the ones that have been described on a genetic basis [30].
Bacteriophages, such as Siphoviridae, are involved in the modulation of bacterial communities and therefore potentially influencing health outcomes [13,31]. In our cohort, the presence of Siphoviridae was extensively negatively correlated with both antibacterial DNA and antiviral cytokine immune responses, while there was a positive correlation of inflammation-related cytokines (IL6, IL8, TNF) following LPS stimulation. It is probable that a robust antimicrobial capacity limits the potential of bacterial growth, consequently reducing bacteriophage proliferation [32]. This finding may have important implications, as it suggests a potential role of bacteriophages as sensitive sensors of host immunity. Although data are scarce, the effect of bacteriophages on the immune system appears to be mostly indirect, through their impact on their target bacteria [13,19,31]. Nevertheless, more complex viral-bacterial interactions may contribute to these observations [13].
It is more challenging to explain the observed positive associations between the presence of anelloviruses and mostly inflammatory (IL6, MIP1b) and regulatory (IL10, IL27) cytokines following bacterial, and to a less extent also viral stimulation. Anelloviruses are apparently non-pathogenic viruses that have been associated with conditions of immune suppression [33] and are considered an integral part of the respiratory virome, particularly in asthma [5]. Anellovirus-dominated profiles as well as anellovirus presence were equally distributed among both antibacterial and antiviral immune response clusters, suggesting that anellovirus presence may be controlled by mechanisms other than TLR-stimulation. It is possible that inflammatory instead of antiviral responses may facilitate anellovirus proliferation.
When studying the microbiome, it is challenging to differentiate the causal host-microbiome associations from secondary microbial changes. In many immune-related conditions, abnormal viral-bacterial interactions can be considered as either a cause or a marker of the disease state [34, 35]. Our results highlight important correlations between the respiratory virome and immune signatures, however, we cannot establish causality. A noteworthy observation is that even though the cohort was comprised of healthy and atopic asthmatic individuals, disease was not a modifier of the correlations, suggesting a fundamental mechanism of immune-microbial interaction; or the result of insufficient statistical power to identify such patterns.
One limitation of the study is the measurement of one post-induction time point, so we could only describe cross-sectional associations instead of a complete response curve, due to practical limitations. The number of samples was moderate, however, we used robust and state-of-the-art methodology for cytokine measurements as well as for the characterization of the virome. Subjects from a wide geographic representation were included. All participating centres followed a validated and synchronized approach for PBMC cultures with common training and reagents. The processing and assessment for nasopharyngeal samples and immune responses was independent.
In conclusion, there are tight parallels between the upper airway virome and the host immune status and potential innate immune responses. Viral stimulation has the capacity of directing immune responses, while immune responses themselves may control microbial composition. The unravelling of such interactions offers new opportunities for intervention towards disease prevention.