In this observational study, we selected 110 biobanked plasma samples, corresponding to all first available samples, i.e. closest to the date of entry in a large open cohort of PLHTLV-1, corresponding to a total of 946 person-years of clinical follow-up, including 43 HAM/TSP patients and 67 asymptomatic PLHTLV-1. Using these unique samples, we demonstrated systemic cytokines and GlycA as candidate biomarkers of inflammaging, immunopathogenesis and therapeutic response in HAM/TSP.
Inflammaging has been extensively documented in people living with HIV-1 (14.15), but this is the first report of inflammaging in PLHTLV-1, characterized by an age-dependent increase in pro-inflammatory cytokine IL-6, which was positively correlated to chronic inflammation marker GlycA (Fig. 2). Among pro-inflammatory cytokines, IL-6 uniquely predicts global functional decline in aging [28] and inflammaging in a systematic review and meta-analysis [29]. Although IL-10 was weakly correlated with age in asymptomatics, the IL6/IL10 ratio, representing a higher pro-inflammatory state, significantly increases with age (± 50%/15years, p = 0.014), and also corroborates the increased mortality rate we observed in this cohort [7]. Since IL6/IL10 ratio has been demonstrated as a sensitive biomarker of COVID-19 outcome [30], aging PLHTLV-1 might be at increased risk of developing severe or critical COVID-19. Although this hypothesis was also supported by in silico findings [31], it remains to be confirmed in large observational studies, which are challenging in neglected diseases [32].
Regarding disease status, we have used complimentary analytical approaches, namely multivariable regression, machine learning-derived decision trees and Bayesian network learning. While multivariable logistic regression identified IL-17A and proviral load as independent predictors of HAM/TSP disease status, Bayesian network analysis enables visualization of the co-dependencies between cytokines in PLHTLV-1, and their associations with GlycA and disease status. Thus, we found that IL-17A appears intrinsically related to disease status in all three analytical models (regression, decision tree and Bayesian network). However, IFN-γ appears “upstream” of all other cytokines in the Bayesian network, which corroborates previous transcriptomic findings [27]. This study confirms the findings of Kagdi et al. for IFN-γ and IL-17A as biomarkers of untreated HAM/TSP [23]. In spite of our larger cohort (67 AS, 43 HAM/TSP), we did not observe increased IL-2 nor IL-10 in HAM/TSP, in contrast to Kagdi et al. (17 AS, 28 HAM/TSP). However, our decision tree (Fig. 2B) identified by Machine Learning is quite similar to the decision tree proposed by Kagdi et al. to classify AS, HAM/TSP and ATL patients based upon IL-10 and IL-17 levels. Exacerbated IFN-γ production has been consistently demonstrated by numerous groups, either ex vivo (in serum/plasma) or in vitro (in supernatants of PBMCs cultured for 1–4 days), as a hallmark of HAM/TSP [4, 5, 23, 33–38]. Other pro- and anti-inflammatory cytokines (IL-2, IL-4, IL-6, IL-10, IL-17A, TNF) have yielded strongly diverging results between different groups regarding their up- or down-regulation in HAM/TSP vs. AS [23, 33–40]. This is most likely due to smaller cohort sizes, in addition to the choice of plasma/serum vs. cell culture supernatants or intracellular flow cytometry. Additionally, differences in age, gender, genetics as well as the inclusion of treated HAM/TSP patients in some cohorts might also explain the observed discrepancies. Indeed, we found that IFN-γ and IL-17A are differentially impacted by corticosteroid pulse therapy: post-treatment IFN-γ levels are low in responders, while IL-17A levels decrease uniformly for all patients. Current clinical guidelines for HAM/TSP suggest that early HAM/TSP patients might benefit most from corticosteroid therapy [6], which is also supported the recent (and first placebo-controlled) randomized clinical trial for corticosteroid therapy in HAM/TSP (HAMLET-P [41]).
In addition to IL-17A, we found proviral load is an independent biomarker of untreated disease in HAM/TSP patients, consenting with the literature [4–7]. However, proviral load did not predict incident HAM/TSP cases in three out of four published Brazilian cohort studies [42–45]. Of those, only Tanajura et al. demonstrated proviral load as a significant predictor of neurological symptoms, but not definite HAM/TSP, during clinical follow-up [43]. Similar to Yamauchi et al. [8], we found that proviral load is not a biomarker for therapeutic response in HAM/TSP. However, we identified, for the first time, TNF and GlycA as independent predictors of clinical worsening, as measured by increased Osame Motor Disability Scale. To put this prediction into a patient-centered clinical context, the median Osame Motor Disability Scale of 4 in the low TNF/GlycA group corresponds to “needs a handrail when climbing stairs”, whereas the median of 8 in the high TNF/GlycA group corresponds to “can walk 1-5m with bilateral support”. Notably, the high TNF/GlycA group also comprised the only fatal case among 43 HAM/TSP patients, with death related to HAM/TSP as described in our previous study [7].
Notable strengths and limitations of this study merit further detail. First, this study has a relatively large sample size, to our knowledge the largest yet with regards to plasma cytokines as candidate biomarkers in PLHTLV-1 and HAM/TSP. Second, complete neurological evaluations of PLHTLV-1 and uniform treatment strategy for HAM/TSP patients are major strengths, as well as a remarkably long follow-up (median 8.6 year in Brazil cohort, > 14 year in US cohort). A major limitation is the low incidence in the US cohort (2/2100 person-years), thus limiting our statistical power for replication. Other limitations include the lack of simultaneous protein and RNA quantification in both cohorts (due to sample availability), as well as potential selection biases regarding patient recruitment and loss to follow-up [7], which are inherent to cohort studies in neglected diseases.