Objectives A population of atypical memory B cells (AtMs) are greatly expanded in patients with active lupus, but their generation and pathophysiological roles are poorly defined. The aim of this study was to comprehensively characterise lupus AtMs with a purpose to identify therapeutic clues to target this B cell population in lupus.
Methods Peripheral B cell subsets were measured by flow cytometry. Sorting-purified B cell subsets were subject to RNA sequencing and functional studies. Plasma cytokines and secreted immunoglobulins were detected by Luminex or ELISA. In situ renal B cells were detected by multiplexed immunohistochemistry.
Results CD24−CD20hi AtMs were strongly increased in two Chinese cohorts of patients with treatment-naïve lupus. Gene expression profile indicated that B cell signalling and activation, lipid/saccharide metabolism and endocytosis pathways were abnormally upregulated in lupus AtMs. In addition, the mammalian target of rapamycin complex 1 (mTORC1) pathway was remarkably activated in lupus AtMs, and blocking mTORC1 signalling by rapamycin abolished the generation of T-bet+ B cells and terminal differentiation of lupus AtMs. Furthermore, lupus AtMs displayed a dysfunctional phenotype, underwent accelerated apoptosis, poorly co-stimulated T cells and produced proinflammatory cytokines. Interestingly, lupus AtMs were in a paradoxically differentiated status with markers pro and against terminal differentiation and enriched with antinucleosome reactivity. Finally, AtMs were accumulated in the kidneys of patients with lupus nephritis and associated with disease severity.
Conclusions These findings demonstrated that mTORC1-overactivated lupus AtMs are abnormally differentiated with metabolic and functional dysregulations. Inhibiting mTORC1 signalling might be an attractive option to target AtMs and to improve therapeutic effectiveness in patients with lupus.
- systemic lupus erythematosus
- B cells
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
What is already known about this subject?
T-bet+CD11c+ atypical memory B cells (AtMs) are greatly expanded in lupus and implicated in the pathogenesis of systemic lupus erythematosus (SLE).
What does this study add?
We find that the mTORC1 pathway is highly activated in lupus AtMs and plays a critical role in the generation and terminal differentiation of these cells. Importantly, inhibiting mTORC1 signalling by rapamycin blocks T-bet+ B cell generation and the terminal differentiation of AtMs.
Our study also demonstrates that AtMs are aberrantly differentiated, metabolically abnormal and functionally dysregulated. In addition, the abundance of AtMs both in the blood and in the kidneys of patients with lupus nephritis reflects the disease activity, indicating that AtMs could be a novel biomarker for patients with lupus.
How might this impact on clinical practice or future developments?
Targeting AtMs by inhibiting mTORC1 signalling pathway could be an attractive option to improve therapeutic effectiveness in patients with SLE.
Systemic lupus erythematosus (SLE) is an autoimmune disease triggered by loss of self-tolerance and the resultant autoreactive cellular and humoral immune responses, leading to a striking heterogeneity of clinical manifestations and organ dysfunction.1 2 B cell abnormality plays a central role in the development of SLE by contributing to the overproduction of autoantibodies, cytokines and augmented presentation of autoantigens to T cells.3 4
The B cell compartment is highly distorted in patients with active SLE. Transitional B cells and plasmablasts are greatly increased, while non-switched memory B cells are decreased in active SLE.5–7 In addition, a population of atypical memory B cells (AtMs) sharing similar phenotypes and identified as CD19hi,8 CD19hiCD21lo,5 9 CD27−IgD− 10 11 are expanded in active SLE. Increased frequency of these AtMs has been associated with high disease activity and disease-specific autoantibodies such as anti-Smith (Sm) antibody,8 10 suggesting that these cells are associated with disease development. Phenotypically similar B cells were also expanded in other autoimmune diseases like primary Sjoegren’s syndrome, systemic sclerosis5 and infectious diseases like HIV infection,12 hepatitis C virus infection13 and malaria.14 Thus, an environment of chronic inflammation seems to promote the generation of these B cells.
Recent studies have revealed that human AtMs specifically express transcription factor (TF) T-bet and integrin CD11c15–17 which are closely linked to a population of B cells named age-associated B cells (ABCs) in aged female mice and young lupus-prone mice.18 19 Murine ABCs are T-bet+CD11c+ and enriched with antichromatin autoantibodies which play an essential role in lupus development.16 19 Furthermore, T-bet is necessary and sufficient for the generation of murine CD11c+ ABCs20 and deletion of T-bet in B cells causes a loss of CD11c+ ABCs and the amelioration of disease in lupus-prone mice.16 21 These studies demonstrated a common pathogenic role of T-bet+CD11c+ B cells in both human lupus and murine lupus models.
The identification of a population of atypical memory T-bet+CD11c+ B cells in patients with SLE raises the possibility that targeting this population may open new therapeutic opportunities for this complex disease, as the efficacies of B cell depletion therapy by targeting whole B cells in patients with SLE are highly variable.1 22 Particularly, it was reported that an expanded population of CD19hi AtMs in patients with SLE predicts a poor clinical response to rituximab treatment.8 Although several factors including Toll-like receptor (TLR) 7, interferon gamma (IFN-γ), interleukin (IL) 21 have been implicated in the differentiation of AtMs,19 20 23 24 no clinically targetable pathway has been identified so far. Thus, to identify the key pathway leading to their generation and to develop feasible strategies to block their development represent an urgent need in lupus B cell study.
In this study, we aim to perform a comprehensive study of lupus AtMs at molecular, cellular, functional and clinical levels, with a purpose to uncover the unknown nature of AtMs and provide clues to target these cells in clinics.
AtMs are increased in patients with lupus and correlated with disease activity
We have analysed peripheral B cell subsets in a Chinese population of patients with new-onset treatment-naïve SLE by flow cytometry and observed that a population of CD24−CD20hi B cells was significantly increased in patients with SLE compared with patients with new-onset rheumatoid arthritis (RA) and healthy donors (HDs) (mean±SEM, 7.54%±0.87%, range 0.19%–25.3%; 3.07%±0.45%, range 0.67%–10.6%; 1.56%±0.15%, range 0.39%–3.68%; respectively) (figure 1A,B) (Detailed characteristics of the three groups were listed in online supplementary table 1). Phenotypical analysis revealed that the CD24−CD20hi B cells expressed high levels of CD11c, T-bet and memory marker CD95, but didn’t express CD21, while the expression of the classical memory marker CD27 was variable (figure 1C), indicating that the CD24−CD20hi B cells were corresponding to a population of AtMs identified in SLE10 16 17 as well as in chronically infectious diseases12 14 and the healthy elderly.25
We then assessed the clinical relevance of these lupus AtMs. We found that the frequency of circulating AtMs was positively correlated with Systemic Lupus Erythematosus Disease Activity Index (SLEDAI) Scores, titres of serum antinucleosome antibody (ANUA) and anti-dsDNA, while negatively correlated with serum complement 3 (C3) and C4 (figure 1D and online supplementary figure 1A). We further showed that the frequency of AtMs was much higher in lupus nephritis (LN) compared with those patients without renal involvement (figure 1E), and was negatively correlated with blood haemoglobin concentration (online supplementary figure 1B). Furthermore the frequency of lupus AtMs was positively correlated with a panel of proinflammatory cytokines including IL-6, IL-18, IFN-α and IFN-γ in plasma (online supplementary figure 1C). We confirmed the correlation between AtMs and disease activity in a second cohort of 19 patients with new-onset lupus (online supplementary table 2). Either defined as CD24−CD20hi or CD11c+CD24-CD20hi, the frequencies of both AtMs were positively correlated with SLEDAI Scores and negatively correlated with serum C3 concentrations (online supplementary figure 1D–F).
Twenty-three patients from the first cohort who have received standard therapy were followed up for 24 weeks (online supplementary tables 3 and 4). The frequencies of AtMs were rapidly decreased 4 weeks after treatment and maintained at relatively low levels thereafter (figure 1F). Strikingly, we observed a significant increase of AtMs in a subgroup of eight patients who experienced relapses or showed resistance to the treatment at week 24 (figure 1G). These results suggest that AtMs could be a potential biomarker to monitor disease activity and predict disease flare.
Lupus AtMs exhibit a unique gene expression profile with BCR activation and metabolic dysregulation
To explore the molecular features of lupus AtMs, we performed an RNA-seq-based gene expression profiling of AtMs and classical memory B cells (CMs) from six patients with new-onset SLE, as well as CMs from five HDs. Principal component analysis separated the three memory B cell populations apart (figure 2A). Venn diagrams showed that 578 genes were significantly upregulated and 384 genes were significantly downregulated in lupus AtMs compared with lupus CMs and healthy CMs, respectively (figure 2B). Gene clustering analysis classified the differentially expressed genes into two major clusters: lupus AtMs high and low expressions (figure 2C). With gene ontology terms as the reference, gene set enrichment analysis (GSEA)26 identified three pathways were predominantly upregulated in lupus AtMs compared with CMs: B cell signalling and activation, lipid and saccharide metabolism and endocytosis (figure 2D,E and online supplementary figure 2A). By contrast, RNA/protein synthesis-related pathways were dampened in lupus AtMs (online supplementary figure 2B).
Consistent with the above pathway analysis, we observed that lupus AtMs displayed heightened B cell receptor (BCR) signalling (pSyk(Y348), pSyk(Y352), pBTK(Y551/511) and pPLCγ2(Y759)) compared with lupus CMs at steady state (figure 2F,G). However, following BCR cross-linking, lupus CMs rapidly upregulated phosphorylated BCR signalling molecules while limited increase was observed in AtMs (online supplementary figure 2C). This BCR responding feature is similar to that observed in the AtMs from malaria-infected patients27 which indicates that AtMs are not intrinsically highly responding to BCR stimulation.
mTORC1 pathway is highly activated and implicated in the differentiation of AtMs
The increased BCR signalling in lupus AtMs at steady state prompted us to check the downstream signalling pathways. GSEA and gene differential analysis showed that the PI3K pathway was mobilised in lupus AtMs (figure 2D, online supplementary figure 3). Strikingly, using the HALLMARK database, GSEA identified that mTORC1 pathway, a key target of the PI3K/Akt pathway, was highly activated in lupus AtMs compared with CMs (figure 3A). Consistently, we observed increased phosphorylations of mTOR(Ser2448) and S6(Ser235/236) in lupus AtMs, indicating activation of the mTORC1 pathway28 29 in these cells. By contrast, the phosphorylation level of Akt(Ser473), a marker of mTORC2 activation,28 didn’t differ between AtMs and CMs in patients with SLE (figure 3B,C).
Next we investigated whether the mTORC1 pathway is involved in the differentiation of T-bet+ AtMs. We chose an in vitro differentiation protocol by using IFN-γ and a TLR7 agonist R848, as both IFN-γ and TLR7 signalling were implicated in the differentiation of T-bet+ AtMs.19 23 30 The results showed that the combination of IFN-γ and R848 had a synergistic effect on mTORC1 activation reflected by the increased frequency of p-S6+ (phosphorylated S6(Ser235/236)) cells and the generation of CD11c+T-bet+ and CD11c-T-bet+ B cells (figure 3D–F). Interestingly, nearly all CD11c+T-bet+ and CD11c-T-bet+ B cells were p-S6+ while the frequency of p-S6+ cells within CD11c−T-bet− B cells was significantly reduced (online supplementary figure 4A,B). The results suggest that T-bet expression in B cells could be associated with mTORC1 activation.
To further explore this connexion, we tested the effect of Rapamycin, a specific inhibitor of mTORC1,31 on the differentiation of T-bet+ B cells. Consistent with an early report,32 Rapamycin decreased the viability and blocked the proliferation of IFN-γ/R848-stimulated B cells (online supplementary figure 4C-F). In gated live B cells, rapamycin strongly decreased the frequencies of p-S6+, as well as CD11c+T-bet+ and CD11c−T-bet+ B cells (figure 3E). Rapamycin also strongly inhibited T-bet expression at mRNA level in stimulated B cells (figure 3F). To differentiate whether the inhibition of mTORC1 activation and T-bet expression by rapamycin is due to the inhibition of proliferation, we focused on non-proliferating B cells. The results showed that the phosphorylation of S6(Ser235/236) and T-bet expression were not dependent on proliferation and rapamycin strongly decreased the frequencies of p-S6+ and T-bet+ B cells in gated live non-proliferating B cells (online supplementary figure 4G,H). Collectively, these studies reveal a potential link of mTORC1 activation and the differentiation of T-bet+ B cells.
Lupus AtMs are enriched with antinucleosome reactivity despite impaired terminal differentiation
There have been inconsistent reports for the terminal differentiation potentials of AtMs, with evidence of normal17 33 or resistant12 27 34 to plasma cell differentiation. Our RNA-seq data showed that lupus AtMs expressed higher levels of B cell identity genes including PAX5, BCL6 and AICDA, and higher amounts of plasma cell-related genes like IRF4, XBP1 and SLAMF7 (figure 4A). At the protein level, AtMs expressed high levels of PAX5, BCL6 and IRF4 (figure 4B). PAX5 and BCL6 are TFs to maintain B cell identity while IRF4 is considered to promote plasma cell differentiation,35 thus AtMs seem to be in an aberrant state of differentiation.
Then we tested the terminal differentiation capacity of sorted lupus AtMs and CMs with a strong stimulating protocol (cytosine-phosphate-guanidine (CpG) 2006/IL-2/IL-10) for 7 days. We found that lupus AtMs displayed a compromised capacity to differentiate into CD27hiCD38hi plasmablasts (figure 4C,D) and to secrete immunoglobulins compared with CMs (figure 4E). Interestingly, after normalisation to the same concentrations, the secreted IgG from AtMs showed an enrichment of ANUA, but not anti-dsDNA antibodies compared with that of CMs (figure 4F). It was reported that mTORC1 signalling was important for plasma cell differentiation and antibody secretion;32 36 we found that rapamycin significantly blocked plasmablast generation, IgG and ANUA production from stimulated AtMs (figure 4G–J). Thus, the above studies indicate that lupus AtMs represented a unique B cell population aberrantly differentiated and enriched with antinucleosome reactivity. Furthermore, inhibiting the mTORC1 pathway could effectively block their terminal differentiation and the capacity to secrete autoantibodies.
Lupus AtMs are molecularly and phenotypically dysfunctional
We observed a panel of genes encoding inhibitory receptors (eg, PDCD1, FCRL3, ILIRB1, ILIRB2) linked to dysfunction or exhaustion in T cells12 27 were highly expressed in lupus AtMs (online supplementary figure 5A), and the dysfunctional phenotype was confirmed at the protein levels for most markers detected (online supplementary figure 5B). Furthermore, markers of B cell development and activation also showed aberrations at molecular and protein levels (online supplementary figure 5A and 6).
TFs play a decisive role in coordinating cell fate and functions, however, the TF network of AtMs is not clear. We found that lupus AtMs expressed a similar TF gene profile to dysfunctional T cells,37 38 including increased expression of BATF, FOXO1, IKZF2, NFATc2, MAF and decreased expression of MAML2, ZEB2, ZBTB20 and TCF7 (online supplementary figure 5C). The enhanced expressions of NFATc2, BATF, FOXO1 and HELIOS were confirmed by flow cytometry (online supplementary figure 5B). Interestingly, we also found that lupus AtMs expressed higher levels of HIF1α and c-Myc (online supplementary figure 5B), two TFs associated with T cell dysfunction39 40 and metabolism.41 42 Thus, the dysfunctional phenotype of lupus AtMs could be linked to a unique metabolic status.
Lupus AtMs are apoptosis-prone, defective for proinflammatory cytokine production and poorly co-stimulate T cells to proliferate
Next we investigated the survival capacity of lupus AtMs. GSEA identified an apoptotic gene signature for lupus AtMs (figure 5A) and several antiapoptotic genes BCL2, PIM1, PIM2 and PIM3 were all decreased in AtMs compared with CMs (online supplementary figure 7A). The downexpression of BCL2 was confirmed at the protein level in AtMs by flow cytometry (online supplementary figure 7B). Consistent with these data, we found that the sorted lupus AtMs displayed significantly less survival advantage and higher frequency of active caspase-3-positive cells than CMs following overnight culture (figure 5B,C).
The RNA-seq data showed that lupus AtMs expressed lower proinflammatory molecules IL6 and Tumor Necrosis Factor ( TNF) and higher anti-inflammatory molecules IL10 and Transforming Growth Factor Beta-1 ( TGFB1) (online supplementary figure 7C). We found that lupus AtMs showed much reduced capacity to produce proinflammatory cytokine IL-6 and TNF-α following polyclonal stimulations (figure 5D and online supplementary figure 7D).
B cells also play an active role in co-stimulating T cells.4 Compared with lupus CMs, AtMs exhibited a significantly decreased capacity to co-stimulate CD4+ T cells to proliferate, as checked by carboxyfluorescein succinimidyl ester (CFSE) dilution (figure 5E), which was accompanied with a diminished survival of AtMs (figure 5F). These results suggest that lupus AtMs are apoptosis-prone, defective for proinflammatory cytokine production and poorly provide co-stimulatory signals to T cells under the current experimental conditions.
AtMs are heavily infiltrated into the kidneys of patients with LN
Finally we performed an in situ study of AtMs in renal tissues of patients with LN by multiplexed immunohistochemistry (figure 6A and online supplementary figure 8A). The renal tissues from patients with LN (detailed characteristics are in online supplementary table 5) showed a tendency of increased infiltration of CD20+ B cells compared with controls (online supplementary figure 8B). Interestingly, CD20+T-bet+ AtMs were abundantly detected in the renal tissues of patients with LN compared with controls for both frequency and density (figure 6B). Consistent with the data from peripheral AtMs, we observed that T-bet+ AtMs from renal tissues of patients with LN displayed significantly higher levels of phosphorylated S6(Ser235/236) than T-bet- B cells (figure 6C,D).
The frequency of AtMs in renal tissues was positively correlated with that in peripheral blood (figure 6E), suggesting that peripheral AtMs frequency may predict the degree of renal AtMs infiltrations. Furthermore, the frequency of renal AtMs was significantly higher in a group with high renal SLEDAI Scores (12~16) than the group with low renal SLEDAI Scores (0~4) (figure 6F). When different pathological types were taken into consideration, type IV LN showed a significantly higher frequency of renal AtMs than that of type III+V (figure 6G). Moreover, the density of renal AtMs was positively correlated with SLEDAI Scores, serum anti-dsDNA titres, urine proteins and renal function indicators while negatively correlated with serum C3 and albumin levels (online supplementary figure 8C,D). These results clearly demonstrated that the degree of renal AtM infiltration is closely associated with disease activity and renal dysfunction, hinting that infiltrated AtMs may directly play a pathogenic role in situ.
The identification of AtMs with a phenotype of CD19hiCD21lo/ − in lupus can be dated back to 2002.9 Since then, a series of studies have reported phenotypically similar B cells to be expanded in patients with lupus.5 8–11 17 23 33 By integrating phenotypical, molecular, functional and histological approaches, the current study adds new evidence that AtMs are abnormally differentiated with metabolic and functional dysregulation, and associated with disease activity. Particularly, we have identified that the mTORC1 pathway is highly activated in lupus AtMs and potentially involved in their generation and terminal differentiation.
In mice, it has been shown that ABCs (murine AtMs) arose from follicular B cells18 and a lot of factors contributed to their generation, including TLR7, IFN-γ, BCR, IL-21, CD40 and major histocompatibility complex (MHC) II molecules.19 20 23 24 Similarly, studies in humans also implicated TLR7/TLR9, IFN-γ and IL-21 in the generation of T-bet+ AtMs-like B cells.17 23 33 However, to our understanding, the key signalling pathway leading to their generation has not been unveiled. Interestingly, we identified that the mTORC1 pathway was particularly elevated in human lupus AtMs and blocking this pathway inhibited the differentiation of T-bet+ AtMs-like B cells. Furthermore, the mTORC1 pathway is known to be critical for plasma cell differentiation and antibody production.43 We confirmed this point by showing that rapamycin strongly decreased the antibody secretion from stimulated lupus AtMs. Given the promising results from a few clinical trials to use rapamycin (sirolimus) to treat patients with lupus,44 45 our current study hints that some effect of rapamycin could be via inhibiting the generation and terminal differentiation of AtMs.
AtMs expressing a panel of dysfunctional/inhibitory markers have been reported in HIV12 and malaria27 infections, as well as in patients with lupus.17 33 We extended these studies by demonstrating that lupus AtMs expressed dysfunctional markers, and shared a TF profile similar to dysfunctional T cells.37 38 Interestingly, lupus AtMs also upregulate the two metabolism-related TFs HIF1α and c-Myc,41 42 which may be linked with the upstream mTORC1 activation and general metabolic aberrations, raising an important question to be addressed in future study.
There is some inconsistence about the capacity of AtMs to undergo terminal differentiation. Several studies have shown the compromised capacity of AtMs to undergo terminal differenation12 27 34 and we obtained similar results. By contrast, in one lupus study, following the stimulation of TLR7/IL-21/IL-10, although lupus AtMs did show a proliferation defect compared with CMs, they produced comparable amount of IgG on a per cell basis.33 Using activated T cells as the stimulus, Wang et al found similar frequencies of antibody-secreting cells and IgG secretion from lupus AtMs and CMs.17 While the difference from the above studies can be ascribed to the different experimental conditions, it should be mentioned that lupus AtMs are in a paradoxically differentiated status: AtMs express high levels of both B cell identity genes SPI1, MEF2C, PAX5 and BCL6, as well as plasma cell-driving genes IRF4, XBP1 and SLAMF7. Our tentative explanation is that AtMs are in fact dragged by two opposing differentiation forces. Most likely AtMs are not essentially poised to plasma cell differentiation themselves, however, they may be efficiently mobilised into plasma cell pool under certain pathogenic conditions like lupus flare.
Our and other studies have shown that lupus AtMs are highly enriched with antinuclear autoantibodies, including Sm, Nucleosome and Chromatin.8 16 17 19 It’s interesting to ask how these pathogenic antibodies are generated and diversified in vivo. The isotype non-switched lupus AtMs (named activated naive B cells) showed higher IgM mutation rates than those of naïve B cells,46 indicating that a mechanism of antibody mutation is working for non-switched AtMs. For switched AtMs, the IgG mutation rates were lower than those of switched memory B cells from the same patients with lupus.33 Together with lower expressions of CXCR4 and CXCR5 on AtMs, the above evidence suggests that AtMs are a population of extrafollicular B cells independent of germinal centre origin. Intriguingly, we found that lupus AtMs express high levels of AICDA, which encodes activation-induced cytidine deaminase (AID), a molecule critical for somatic mutation and immunoglobulin isotype switching.47 We propose that AtMs are uniquely differentiated from a population of self-reactive naïve B cells48 in the context of chronic stimulation. High expression of PAX5 and BCL6 will ensure B cell identity in order for AID to perform its function and generate autoantibodies with sufficient affinity. In the context of disease flare, lupus AtMs could be abundantly expanded and mobilised into plasma cells to produce high levels of pathogenic autoantibodies.
In summary, the current study reveals new molecular and functional features of lupus AtMs. Particularly, as mTORC1 is required for both the generation and the terminal differentiation of AtMs, targeting mTORC1 pathway may represent an alternative strategy to the current lupus therapies, especially for those patients refractory to conventional regimens.
Details of methods are available in online supplementary materials section.
The authors thank Min Dai, Wei Fan, Fang Du, Chunyan Zhang, Huihua Ding, Qingran Yan, Yan Ye, Suli Wang, Xia Lv and Xinyue Lian for clinical support (Renji Hospital, School of Medicine, Shanghai Jiao Tong University, China), and Huihui Xu, Meiying Zhang and Yi Chen for technical support (Institut Pasteur of Shanghai, Chinese Academy of Sciences, China).
CW, QF and QG contributed equally.
Handling editor Josef S Smolen
Contributors CW and SG performed experiments and analysed data. SS, ML, TF and TL performed experiments. QF, QG, SC, FC, ZC, YL and PH analysed data. XC and YH provided reagents. QG, NS, CB and XZ designed the research. CW, QF and XZ wrote the manuscript.
Funding This work was supported by the Strategic Priority Research Program (No. XDB29030302), Interdisciplinary Innovation Team, Frontier Science Key Research Project (No. QYZDB-SSW-SMC036), Chinese Academy of Sciences; 973 Program (No. 2014CB541904), the National Natural Science Foundation of China (Nos. 31470879, 31770960, 81471596, 8157575, 81771733, 81771737 and 81861138010) and Nantong Science and Technology Programme (No. MS120170105).
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data are available in a public, open access repository. All data relevant to the study are included in the article or uploaded as supplementary information.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.