Our current understanding of the role and contribution of the gut microbiota on adverse outcomes in paediatric patients post-HSCT remains in its infancy. Herein, we report the first study from the UK investigating the microbiota-metabolome axis in 64 patients undergoing HSCT at Great Ormond Street Hospital, a major tertiary centre in London.
Our findings indicate that patients generally exhibit lower alpha diversity at baseline (pre-transplant) compared to healthy controls; most patients will have begun prophylactic antimicrobials at this time point; therefore, one may hypothesise that this is not a surprising finding. A further decrease in alpha diversity post-transplantation was common, and in most cases, microbial diversity did not return to pre-transplant baseline levels during the hospitalisation period, a feature consistent with reports in adult HSCT [5]. The absence of obligate anaerobes including Ruminococcaceae and Lachnospiraceae and the observed domination by a single taxon post-transplantation is likely due to extensive and prolonged exposure to antimicrobials, particularly the high anaerobe coverage received by the current cohort. Piperacillin-tazobactam and meropenem, which were frequently administered in our cohort, have been shown to impact obligate anaerobes in an adult HSCT cohort [39]. Taken together, the data suggests that antibiotic stewardship is a key determinant and therefore amenable for intervention for better microbial functional status and clinical outcome post-transplantation.
Consistent with previous reports in both adults and children, we identified the expansion of facultative anaerobes, including Enterococcaceae, Enterobacteriaceae and Streptococcaceae post-transplantation; adding support to the notion that despite the differences in transplant procedure and drug regimens between various hospital centres, expansion of facultative anaerobes is a salient feature post-HSCT transplantation [27, 36, 39, 40]. This is overall in line with another paediatric study; interestingly, the authors also identified the presence of a high Lactobacillaceae cluster [27], which was not seen in our cohort. The drivers for domination of specific families remain unknown; however, the post-transplant gut environment, due to conditioning regimens, diet and drug administration amongst other factors, is likely a major determinant. Although certain dominations are specific to the current cohort, the dynamics of the microbiota appear similar to that seen in adults. Indeed, several murine studies indicate that germ-free mice have defects in haematopoiesis and antibiotic-treated mice show multilineage repression of haematopoiesis [41, 42]. Staffas et al. recently demonstrated that intestinal microbiota contributes to haematopoiesis post-HSCT via improved dietary energy uptake in mice [43]. Studies to better understand the impact of specific taxon dominance on immune reconstitution in adult and paediatric post-HSCT are clearly warranted.
Plotting the samples in a t-SNE space illuminated a dynamic HSCT microbial landscape. Most individuals showed frequent stochastic movements, with multiple dominations throughout their inpatient stay and no return to their initial state (patient X; Fig. 2a, b). In contrast, a few patients exhibited relatively few transitions, and some returned close to their initial microbiota composition (patient Y; Fig. 2c, d). Similar observations about the lack of obligate anaerobes in the baseline samples and stochastic movements across the microbial landscape were also seen in the autologous patients. It is important to note that patient X received a broad-spectrum antibiotic prior to switching to CST3 and that the apparent instability of their microbiome may be in part due to longer hospitalisation and hence overall health post-transplant. At present, it is unclear why certain individuals return to their initial profile, whereas others do not, which warrants further investigation. Such behaviour may be an intrinsic feature, for example, a more diverse microbiota at the beginning of the treatment may be less susceptible to further modulation. The age of the patient, and therefore the maturation of the gut microbiota at the time of transplant, may also play a role in its resilience to continuous perturbation. Likewise, the observed stability or lack thereof may in part reflect the overall health of the patient post-transplant. Additionally, given that patients were followed up for variable amounts of time, this does not exclude the possibility that patients could have recovered at later time points.
Assuming that there is a single equilibrium value with a functionally optimal microbiota for each individual, an insult, such as a course of antibiotics, may lead to perturbation and thus a shift away from this optimum. It is tempting to hypothesise that with the number of continuing insults during HSCT, this single equilibrium value is gradually shifted away to a new value, and eventually, the landscape itself is altered [44, 45]. Palleja et al. showed that healthy adults are resilient to a short course of broad-spectrum antibiotics, as they return to near-baseline composition within 1.5 months [46]. Despite this, certain taxa remained undetectable 6 months post-treatment. Significant antimicrobial and immunosuppressant usage during HSCT have been found to have a profound impact on the microbial landscape [44, 45]. Given the high anaerobe cover in this cohort, it is not surprising that most patients in this study exhibited a perturbed microbial composition on admission and during treatment and did not return to their initial pre-transplant microbiota status during the observation period. Investigations into the optimal use of antimicrobials with a view of preserving the GI microbiota in this population are essential, and a study is already underway in an adult cohort (NCT03078010).
To identify the distinct bacterial community patterns within the population, the data were partitioned into three CSTs, each with a varied microbial composition. Transition models revealed time-dependent patterns such as transitions between CST2 and CST3 to CST1 becoming less common with time, with no such transitions observed after week 3. This suggests that there may be a critical time period after transplantation, during which the microbiota is able to return towards a healthier state. The microbiota around this period may also be more amenable to intervention. Both CST2 and CST3 appear to have high self-transition probabilities, making them more stable than CST1. This is similar to another study in adults, which suggests that the resilience of a biodiverse state is low during the post-HSCT period, with an observed self-transition probability of 49%, in comparison with our observation of 40% [5].
Importantly, we found that CST3 was associated with a higher risk of viraemia after transplantation. CST3 is a state with a complex composition, and on further analysis, we were unable to identify specific taxa responsible for this association, although Enterobacteriaceae was close to significance. A small sample size might be at play, or a higher risk of viraemia could be a result of a cumulative effect of several taxa. The composition of CST3 is complex and includes several Proteobacteria genera including Klebsiella, Escherichia and Enterococcus. It is unclear how Enterococcus and/or other taxa may contribute towards an increased risk of viraemia; however, this could be through indirect action via delayed immune re-constitution. Given the associations between the HSCT gut microbiota and immune cell dynamics, particularly, between white blood cells and anaerobes such as Ruminococcus, Fecalibacterium and Akkermansia, it is plausible that HSCT gut microbiota may impact immune reconstitution and, in turn, viraemia [47]. Additionally, the taxa may act indirectly through reactivation and/or delayed/impaired response to viral infection/reactivation. The commensal microbiota (through MAMPs engaging pattern recognition receptors) as well as derived metabolites are known to impact and generate optimal innate and adaptive immune responses important for controlling systemic viral infections as well as having an impact on viral-specific CD8 T cell memory in a murine model infected with CMV [48, 49]. Additionally, CST3 could simply be a marker of ‘poor’ gut health such as damage to the colonic mucosa, which may have an impact on viraemia development. Several recent publications link CMV reactivation to antimicrobial use in adult HSCT populations. Zhang et al. found an increased risk of CMV reactivation with the use of vancomycin, whereas Camargo et al. found the use of anaerobic antimicrobials in the first 2 weeks post-transplantation associated with a twofold risk of clinically significant CMV reactivation [50, 51]. Given the high use of vancomycin in this cohort, the impact of the microbiota and/or antimicrobials on viral reactivation and clinically significant viraemia would be of interest in the future.
Variations in microbial metabolites were also recorded during HSCT. This included reduced amounts of SCFA and BCFA butyrate and isobutyrate in the baseline patient samples versus healthy controls. Butyrate at baseline positively correlated to microbial diversity and was associated with a reduced risk of viraemia. Although the link to diversity has been previously observed, the association to the risk of viraemia is a novel finding [7, 24, 52, 53]. Butyrate exerts multi-fold effects on intestinal epithelial cell integrity and immune cell homeostasis; thus, higher levels may, to an extent, be indicative of a ‘healthier’ microbiota. This cohort may benefit from an intervention that aims to restore levels of SCFA either directly or through prebiotics, probiotics or postbiotics. Despite this, the administration of butyrate in in vitro models inhibits colonic stem cells from forming an intact epithelial monolayer; therefore, investigations into the most appropriate approaches for this cohort are warranted [54]. In contrast, glucose was significantly higher in the patient group. This may be an indication of a degree of gut damage/malabsorption and/or the result of the expansion of certain facultative anaerobes, specifically Enterococcus [55]. A recent study highlights that lactose drives Enterococcus expansion in a murine model; thus, an increase in glucose may be the result of lactose metabolism; however, either of these hypotheses require further investigation [56].
sPLS and CCA revealed potential associations between the three clusters and patient clinical parameters. Of interest, cluster 1, which was high in obligate anaerobes, was associated with the non-use of TPN and the use of cotrimoxazole and piperacillin. Detailed effects of TPN on GI microbiota have not been elucidated, although the use of TPN and higher GvHD incidence and worse survival have been reported [53]. There is an ongoing debate on whether enteral nutrition may be preferable over TPN for preserving the GI barrier integrity and ecology. The association with cotrimoxazole and piperacillin is less clear given that piperacillin is a broad-spectrum antimicrobial and has been shown to have an effect on obligate anaerobes [39]. Cotrimoxazole is a PCP prophylactic, and previous work has shown that it has a somewhat limited effect on the microbiota composition [57].
Cluster 2 associated with the use of multiple antimicrobials, including metronidazole, vancomycin and azithromycin, all known to be detrimental to the commensal gut anaerobes, likely contributing to the milieu of taxa observed in this particular cluster [46]. Finally, cluster 3 also associated with the use of various antimicrobials including meropenem and ceftazidime. Meropenem and piperacillin have a detrimental impact on obligate anaerobes whereas ceftazidime was found to have a moderate effect in a HSCT cohort [39]. Quinolones (e.g., ciprofloxacin) in contrast exert a marginal effect [39]. Despite this, given that no use of ciprofloxacin was associated with cluster 3, there is likely an effect in this cohort. Interestingly, in agreement with the CST findings, viraemia and TPN both contribute to cluster 3. Additionally, acetate was significantly lower in baseline samples of allogeneic HSCT patients receiving TPN than in those without TPN. Whilst the associations appear rational, further insight into the exact dynamics of antimicrobial administration and the use of TPN in this patient cohort requires a detailed investigation. Furthermore, given the associations between the type of donor and cluster 3, it would be of interest to also profile donor gut microbiota where feasible. The lack of associations between GvHD and the GI microbiota is somewhat surprising given the previous findings; however, it may be explained, to some extent, by the heterogeneity of the cohort, making a signal more difficult to identify.
In summary, the current study investigated longitudinal microbiota and metabolome in paediatric HSCT patients at a single centre. Despite differing conditioning and treatment regimens between transplant centres, several salient features between adult and paediatric HSCT were identified including domination by a single taxon and no return to baseline within the hospitalisation period. Our study is the first to suggest an association between microbiota and butyrate levels at baseline and risk of viraemia. Although our report is the largest longitudinal study in paediatric HSCT to date, given the heterogeneity of the patients, the sample size is relatively small, which prevents us from undertaking more in-depth analysis including the stratification of the patients by their underlying condition or age. We collected weekly samples during the hospital stay; therefore, the microbiota data is limited, and it is plausible that we were unable to fully capture all dominations, which could improve CST resolution. In order to comprehensively profile the microbiota, we sampled patients weekly; however, due to patient heterogeneity, such as discharge times, this led to irregular overall sampling times. It would be of interest to extend to similar sampling times for all patients in the future.
The impact of certain co-variates known to affect the microbiota such as nutrition and antimicrobial use underwent limited investigation; therefore, an in-depth study on their effects is necessary. Patients in our cohort frequently had co-infections; we were therefore unable to delineate the specific associations between specific viraemia, i.e., CMV/EBV and the microbiota. As a result of previous findings, associations between bacteraemia and the microbiota were also of interest in this cohort; however, given the low number of bacteraemias in this cohort, we were unable to interrogate this further.
The samples size of the healthy controls was relatively small in the present study; in addition, the HC were not followed longitudinally. We observed minimal variability in the clustering of the HC metabolomics profiles (unlike the patient samples), which highlights the marginal variability of these samples in comparison with patient samples; it is likely that a larger sample size of HC would be unlikely to provide a greater deal of additional information. Likewise, given that microbiota remains fairly stable once mature, repeat sampling was unlikely to add additional insights to the study. Finally, whilst 16S rRNA sequencing remains a useful way to taxonomically profile the microbiota, in order to gain a deeper understanding of microbial dynamics, future studies must employ metagenomic and metatranscriptomic approaches to gain molecular insights into microbial-host interactions. Despite these limitations, the study provides a reference point for further vital work in this population.