A Plasma miR-193b-365 Signature Combined With Age and Glycemic Status Predicts Response to Lactococcus lactis–Based Antigen-Specific Immunotherapy in New-Onset Type 1 Diabetes

Type 1 diabetes is an autoimmune disease characterized by the selective destruction of the insulin-producing pancreatic β-cells (1). Its heterogeneous presentation and course imply that a “one-size-fits-all” therapy is unlikely to provide the desired cure that can halt disease development in all patients (2). Thus, patient care should shift to precision medicine (3,4), based on the identification of specific patient endotypes with unique genetic, biological, and environmental properties (5).

Immunotherapies targeting T and B cells, using anti-CD3 (i.e., teplizumab and otelixizumab) (6–8) and anti-CD20 (i.e., rituximab) (9) monoclonal antibodies (mAbs), respectively, demonstrated acceptable safety and efficacy in maintaining functional β-cell mass, albeit temporarily, in patients with newly diagnosed type 1 diabetes. Hence, based on initial positive results, teplizumab (6,10,11), antithymocyte globulin (ATG) (12–14), rituximab (9), and abatacept (CTLA4-Ig fusion protein) (15) therapies are further explored in late-stage prediabetes and type 1 diabetes onset, but they still face variable therapeutic outcomes differences in treatment response on an individual basis (16). Furthermore, little information is available on their long-term potential to restore peripheral tolerance or on the type of patient who would benefit most from these therapies.

Combinations of immunotherapies with disease-relevant antigens, delivered via tolerogenic routes, hold the potential of greater disease specificity, lower toxicity, and decreased risk of tumor susceptibility and infections, and are a putative solution for preventing or reversing autoimmunity (17). Our team demonstrated that a short course of low-dose anti-CD3 mAbs combined with an oral delivery of genetically modified Lactococcus lactis bacteria secreting full human proinsulin (PINS) and the anti-inflammatory cytokine IL-10 stably reversed new-onset type 1 diabetes in the nonobese diabetic (NOD) mouse model (18–20). Preclinical success was related to glycemic status and insulin autoantibody titer at study entry, but these disease-associated parameters could not entirely identify which mice would respond to the therapy. A first-in-men study with a clinical L. lactis–based product (AG019 ActoBiotics) combined with teplizumab demonstrated stable C-peptide, HbA_1c, and insulin-dose adjusted A1c values, in addition to an induction of antigen-specific immune responses in patients with newly diagnosed type 1 diabetes (21). These observations and the need to identify accurately those who will benefit from the treatment and those who will likely not respond underscore the exploration for highly specific biomarkers that can provide information not only on β-cell health and function but also on the complex immune signature at baseline.

miRNAs are evolutionarily conserved, small, noncoding RNA molecules that regulate gene expression at the posttranscriptional level in important cellular processes (21–24). miRNAs have been found intracellularly but can also be selectively released from different cell types (25). These extracellular miRNAs can enter the bloodstream either encapsulated in small vesicles (26) or associated to Argonaute 2 complexes (27). They are present in a variety of biological specimens such as peripheral blood, saliva, urine, and feces. These relatively stable molecules hold great potential to be applied to disease diagnosis, monitoring, prognosis, and prediction of therapy response (reviewed by Sebastiani et al. (28)). Notably, certain circulating miRNAs, some of which have been linked to β-cell function and/or glucose homeostasis (reviewed by Dotta et al. (29)), were shown to be dysregulated in plasma or serum from patients with type 1 diabetes (30,31).

In this work, we aimed to identify cell-free circulating miRNAs in plasma of NOD mice with new-onset diabetes at study entry that are able to predict the outcome of the L. lactis–based immunotherapy. We found that an miR-193b-3p and miR-365–3p plasma profile improved the prognostic power of particular clinical parameters for therapeutic benefit. The composite biomarker signature could guide clinical decision-making in patients with newly diagnosed type 1 diabetes as well as accelerate future clinical trial execution.

Mice were bred and housed according to protocols approved by the KU Leuven Animal Care and Use Committee (Leuven, Belgium; project no. P116/2015 and P068/2019), and experiments complied with the European Union (EU) Directive 2010/63/EU for animal experiments. For details, see the Supplemental Material.

Here we briefly describe various assays and analyses we conducted for this study. For additional details, see the Supplemental Material.

Pancreatic islets were isolated from C57BL/6 mice (18–20 weeks old) using intraductal collagenase digestion and then handpicked. Genetically modified L. lactis bacteria secreting the human full PINS antigen and human IL-10 were generated by Precigen ActoBio (Zwijnaarde (Ghent), Belgium) and grown as described (18,19). NOD mice with new-onset diabetes were bled via the submandibular vein and treated for 5 consecutive days with 2.5 μg/d of hamster anti–mouse CD3 mAb (clone 145–2C11; BioXCell, West Lebanon, NH), combined with an oral gavage of L. lactis bacteria secreting human PINS and IL-10 (2 × 10⁹ CFU/d) 5 times per week for 42 days (Fig. 1A).

Total RNA, including miRNAs, was extracted from 50 µL of mouse plasma or from cytokine-treated mouse islets, using an miRNeasy micro kit (Qiagen, Hilden, Germany). For plasma samples, single-assay, stem-loop RT-qPCR was used to validate miRNA expression from TaqMan miRNA array analysis (32).

Single-cell suspensions from NOD pancreas and whole blood were stained and sorted directly into Trizol-LS (Thermo Fisher Scientific, Waltham, MA). Droplet digital PCR (ddPCR) was performed using the QX200 Droplet Digital PCR System (Bio-Rad Laboratories, Hercules, CA) to quantify miR-193b-3p and miRNA-365–3p in flow-sorted immune cells derived from NOD pancreas and whole blood. We performed cellular indexing of transcriptomes and epitopes (CITE) sequencing, as described by Stoeckius et al. (33), with small modifications. Data were analyzed using Seurat, version 4.1.0. To investigate potential relationships between upregulated differentially expressed genes (DEGs) in pancreas-infiltrating neutrophils and basophils, the STRING (version 11.5) tool (https://string-db.org) was used.

All statistical analyses were performed using GraphPad Prism software, version 9.0 (GraphPad, San Diego, CA). For details, see the Supplemental Material.

All data sets generated and analyzed in this study are available from the corresponding authors on reasonable request. All mouse CITE-sequencing raw data and mouse gene-cell count matrices are deposited at Gene Expression Omnibus (National Center for Biotechnology Information) under accession number GSE214031.

Based on a large data set of NOD mice with new-onset diabetes (n = 110; Supplementary Table 1) treated with a combination therapy (CT) of low-dose anti–CD3 mAbs (clone 145–2C11) and genetically modified L. lactis bacteria secreting full human PINS and IL-10, we evaluated the influence of sex, age, and glycemic status at disease onset on the responsiveness to the L. lactis–based immunotherapy (Fig. 1A). The CT-treated group had an overall disease remission rate of 45% compared with untreated diabetic controls (P < 0.01) that remained hyperglycemic and were sacrificed on the basis of human end points (Fig. 1B).

NOD mice develop type 1 diabetes with a heterogeneous presentation and course. Mean (±SD) age and glycemia at disease onset were 16 (±3.4) weeks of age (range, 9–27 weeks) and 19.5 (±5.15) mmol/L (range, 11.7–33.3 mmol/L) respectively, in the studied cohort. Based on this information and previous studies (34), we used 16 weeks of age and 19.4 mmol/L blood glycemia as cutoff values to investigate the impact of age and disease presentation at study entry on therapy outcome. Although we found no significant sex- or age-dependent differences in the outcome of the CT response, mice with mild hyperglycemia (<19.4 mmol/L) at disease onset had a superior response to the CT compared with mice with severe hyperglycemia (>19.4 mmol/L; 65% vs. 29% remission, respectively; P < 0.0001)(Fig. 1C). Receiver operating characteristic (ROC) curves compared sensitivity versus specificity for predicting therapy response, with a range of values from age and glycemia, alone or in combination (Supplementary Fig. 1, Supplementary Table 2). Although age at disease onset by itself did not influence therapy outcome, it significantly increased the predictability of therapy response when combined with glycemic status at disease onset (area under the curve [AUC]: 0.8; sensitivity, 64%; specificity, 86%; P < 0.001).

To investigate circulating miRNAs as potential biomarkers to predict responsiveness to the L. lactis–based immunotherapy, we profiled expression values of 378 miRNAs in plasma samples collected at study entry. A detailed work flow depicts the experimental design and data analysis (Supplementary Fig. 2). The initial profiling was carried out using a TaqMan miRNA array on plasma samples obtained from six responder (R) and six nonresponder (NR) mice to the L. lactis–based immunotherapy, with even numbers of mice having mild (<19.4 mmol/L) and severe (>19.4 mmol/L) hyperglycemia at disease onset in each of the outcome groups. Through this analysis, we identified 236 miRNAs that were consistently expressed in at least one group, and we generated a hierarchically clustered heat map. Overall, we did not observe any significant differences in the number of detected miRNAs (Supplementary Fig. 3). Color-scale representation of the expression values allowed us to identify samples characterized by high or low miRNA expression in plasma (Fig. 2A).

A volcano plot depicting differential expression revealed six miRNAs (i.e., miR-34a-5p, miR-125a-3p, miR-193b-3p, miR-328, miR-365–3p, and miR-671–3p) upregulated at study entry in plasma of NR versus R mice in the discovery cohort (Fig. 2B, Supplementary Table 3). Four of the six miRNAs were further confirmed by single-assay stem-loop RT-qPCR (Fig. 2C). Remarkably, these miRNAs were not overexpressed in plasma collected at study entry in NR compared with R mice given anti-CD3 monotherapy (Supplementary Fig. 4).

Plasma samples collected at study entry from an independent cohort of 54 NR and 44 R mice to the L. lactis–based immunotherapy were used to validate the differentially expressed plasma miRNAs from the discovery phase. miR-125a-3p expression was significantly increased in plasma of NR compared with R mice subjected to the L. lactis–based immunotherapy (Fig. 2D). ROC analysis revealed that miR-125a-3p could predict therapy response (P = 0.02); however, its prognostic power (AUC: 0.6; sensitivity, 65%; specificity, 64%) was inferior to the use of the aforementioned clinical parameters (Fig. 2E).

We hypothesized that a combined signature composed of differentially expressed miRNAs and clinical parameters could improve the prognostic power for therapy response. We stratified mice at study entry on the basis of glycemic status (< or >19.4 mmol/L), age (< or >16 weeks old), or a combination thereof (Fig. 3A), and evaluated miR-34a-5p, miR-125a-3p, miR-193b-3p, and miR-365–3p expression in these subgroups (Supplementary Fig. 5). Although miR-125a-3p in combination with younger age at study entry with or without severe hyperglycemia (Fig. 3B and C) had superior prognostic power for therapy response compared with values obtained in the unstratified group (Fig. 3F and G, Supplementary Table 4), miR-193b-3p and miR-365–3p combined with older age and severe hyperglycemia at study entry (Fig. 3D and E) outperformed all other combinations in predicting poor response to the L. lactis–based immunotherapy (AUC: 0.9; sensitivity, 94%; specificity, 75%) (Fig. 3H–J, Table 1, Supplementary Table 4).

Because disease severity seemed to influence the response of NOD mice with new-onset diabetes to the L. lactis–based immunotherapy, we hypothesized that miR-193b-3p and miR-365–3p might originate from either the target organ or cells exposed to (or responsible for) disease-relevant inflammation. To investigate the cellular origin of miR-193b-3p and miR-365–3p, we first studied whether they originated from islet cells under inflammatory stress. We measured their expression in wild-type C57BL/6 mouse primary islet cells exposed to a cytokine mix (i.e., IL-1β plus IFN-γ) for 6 or 24 hours. We found no evidence of cytokine-mediated modulation of miR-193b-3p and miR-365–3p expression (Fig. 4A and B), making islet cells unlikely as a cellular source. These data are in line with previous reports on miRNAs expression in murine and human pancreatic islet cells upon cytokine treatment (35,36), which found no relevant changes in miR-193b-3p and miR-365–3p expression.

To identify potential immune cell subsets that could contribute to the plasma miR-193b-3p and miR-365–3p pool in NOD mice with new-onset diabetes, we sorted by flow cytometry 11 major immune cell subsets (i.e., B cells, T cells, CD4⁺ T cells, CD8⁺ T cells, natural killers cells, γδ T cells, monocytes, dendritic cells, neutrophils, basophils, and myeloid cells) (Supplementary Fig. 6A–D) from both peripheral blood and pancreas isolated from NOD mice with new-onset diabetes (Supplementary Table 5). miR-193b-3p and miR-365–3p expression values were then studied by ddPCR, measuring absolute copies, and expressing the data as a ratio to the standard small nuclear RNA RNU6 (Supplementary Fig. 7). Comparing the expression values of miR-193b-3p and miR-365–3p between peripheral blood and pancreas for each immune cell subset, we found that both miRNAs were more abundant in basophils infiltrating the pancreas than those isolated from peripheral blood (Fig. 4C and D). Remarkably, the same differential expression was observed in neutrophils for miR-193b-3p (Fig. 4C), whereas a trend was noticeable for miR-365–3p (Fig. 4D). When we ranked the immune cell subsets on the basis of the expression of miR-193b-3p and miR-365–3p in peripheral blood and pancreas, we found that basophils and neutrophils infiltrating the pancreas had the highest expression. Blood-circulating neutrophils also showed the highest expression of miR-365–3p compared with expression values in other immune cell subsets (Fig. 4E). Overall, these data suggest that neutrophils and basophils have the highest expression of miR-193b-3p and miR-365–3p in both peripheral blood and pancreas and that their expression was increased when these two immune cell populations were present in the pancreas compared with those from peripheral blood.

Because miR-193b-3p and miR-365–3p expression values were increased in pancreas-infiltrating neutrophils and basophils of NOD mice with new-onset diabetes, we used CITE-sequencing technology to further investigate their phenotype and frequency. CITE sequencing enabled us to perform a multimodal immune cell phenotyping by measuring cell surface proteins as well as the entire transcriptome at the single-cell level using a panel of 190 DNA-barcoded antibodies (Supplementary Table 6). This analysis was carried out on FACS-sorted CD45⁺ leukocytes isolated from peripheral blood and pancreas of four NOD mice with new-onset diabetes (Fig. 5A and B). We obtained a data set of 43,706 cells in total. Unsupervised uniform manifold approximation and projection (UMAP) analyses implemented in the Seurat pipeline were used to identify the major immune cell populations (Fig. 5C). All UMAPs based on RNA and protein showed a clear separation of the identified immune cell types (Supplementary Fig. 8A). The cellular identity of the different clusters was assigned on the basis of the differential expression of hallmark genes and surface proteins (Fig. 5E, Supplementary Fig. 8B, and Supplementary Fig. 9). Most gene–protein combinations had a strong correlation, although some markers (i.e., Ly6G, Itgax, and Fcer1a) showed variable expression in all cell types (Supplementary Fig. 9). The major populations identified were distributed differently in peripheral blood and pancreas. The pancreas was infiltrated primarily by lymphocytes, whereas myeloid-lineage cells predominated peripheral blood (Fig. 5D–F). In fact, the majority of basophils and neutrophils included in the CITE-sequencing analysis were found in peripheral blood rather than in the pancreas of NOD mice with new-onset diabetes. Neutrophils also represent the most abundant peripheral blood–circulating immune cell subset, while basophils contribute to a lesser extent (Fig. 5G).

Using DEG analysis, we further compared the transcriptome profiles of pancreas-infiltrating and blood-circulating neutrophils and basophils of NOD mice with new-onset diabetes. The volcano plots show that many genes in both cell types were upregulated in the pancreas compared with those in the circulation (Fig. 6A and B). According to an in silico coexpression analysis, the majority of these upregulated genes were part of different transcriptional networks of strong protein–protein interactions (PPIs). We found a group of 12 coregulated genes, including the activator protein-1 dimeric transcription factor components (i.e., Jun, Junb, Jund, Fos, and Fosb), members of the dimeric NF-κB signaling pathway (i.e., Nfkbia, Nfkbid, and Nfkbiz), in addition to Zfp36, Egr1, Ier2, and Nr4a1. These transcription factors act as master regulators of inducible gene transcription, modulating inflammatory responses and cell-fate decisions (37). Our data also revealed regulatory networks of the IL-1β signaling pathway in neutrophils and the IL-6 signaling pathway in basophils, both of which are known to be involved in inflammatory processes. Furthermore, neutrophils infiltrating the pancreas expressed high values of Cxcl2, a chemokine known to mediate recruitment and activation of immune cells in inflamed tissues (38). On the other hand, basophils infiltrating the pancreas expressed high values of activation and cytotoxic markers such as CD69 and Gzmb (Fig. 6C and D). A more detailed pathway analysis confirmed that the NF-κB and inflammatory pathways were upregulated in neutrophils and basophils infiltrating the pancreas compared with those in the circulation (Fig. 6E and F). These findings suggest that the plasma values of miR-193b-3p and miR-365–3p may be connected to the inflammatory and activation status of the pancreas-infiltrating granulocytes.

Disease-modifying immunotherapies aimed at re-educating the host immune system to peripheral tolerance and maintaining residual functional β-cell mass have the potential to change the course of type 1 diabetes treatment and may offer the long-elusive prospect of a cure. Although single-agent therapies like teplizumab (6,10) and ATG (13) transiently preserved β-cell function in some individuals, most study participants had ongoing β-cell loss, suggesting that tolerance was not reached. The way forward is the rational design of combination therapies that not only improve therapeutic success with minimal toxicity but simultaneously reduce the likelihood of therapy unresponsiveness in an individual patient (39,40). The identification of patient endotypes based on demographic, genetic, histopathological, immune, and metabolic features will most probably accelerate the implementation of precision medicine and can influence trial design and execution (4).

We developed an experimental immunotherapy composed of a short course of low-dose anti-CD3 mAbs with a L. lactis–based delivery of full human PINS and IL-10. This combination therapy stably and permanently reverted new-onset type 1 diabetes in most, although not all, mice (18–20). A first-in-human study with the L. lactis–based immunotherapy (ActoBiotics AG019) demonstrated, apart from acceptable safety and tolerability, favorable metabolic and antigen-specific immune responses (41). The next step to human translation is to better select those individuals who are likely to respond. Here, we propose that on top of clinical parameters, cell-free circulating miRNAs may support patient stratification but also accelerate future trial execution.

Previous and current data obtained with the L. lactis–based immunotherapy linked glycemic status at study entry to therapy response (19) but only with moderate sensitivity and specificity (70% and 70%, respectively). Thus, additional prognostic biomarkers are needed to improve patient selection and increase the therapeutic success rate. In the present study, using multiple mouse cohorts and optimized standard operating procedures with different validation and stratification steps (31,42), we uncovered circulating miRNAs differentially expressed between specific subgroups of NR and R mice subjected to the L. lactis–based immunotherapy. Although plasma miR-125a-3p expression could discriminate NR from R mice in both the discovery and validation cohorts, its prognostic power for therapy response was not superior when compared with ROC curves and corresponding AUC values when age and glycemic status at study entry were applied. Several reports highlight the role of miR-125a-3p in T-cell biology (reviewed by Wang et al. (43) and Sun et al. (44)). miR-125–3p was induced and highly expressed in T regulatory cells and not in effector T cells. miR-125–3p might affect effector T-cell differentiation. Hence, miR-125a-3p overexpression in plasma of NR mice remains of interest, especially given the T-cell–directed nature of the immunotherapy, even though its utility as a prognostic biomarker for therapy response was relatively low and outperformed by other biomarker signatures. Indeed, the combination of plasma miR-193b-3p and miR-365–3p overexpression, together with age and glycemic status at study entry, had the best prognostic power for therapy response (AUC: 0.9; sensitivity, 94%; specificity, 75%). Importantly, these observations highlight the importance of integrating multiple omics covariates with disease-associated parameters to obtain a robust therapy response signature. In this regard, consortia like INNODIA already drafted master protocols that, apart from metabolic parameters, integrate multiomics analyses on various biological samples (i.e., blood, urine, and feces) in their upcoming clinical trials with the intention to get a broader picture of the underlying pathological mechanisms in their study participants (45).

Although the role of the miR-125 family in the immune system is well addressed (by Wang et al. (43) and Sun et al. (44)), the putative origin and function of miR-193b-3p and miR-365–3p are less clear. miR-193b-3p and miR-365–3p are transcribed from a highly conserved cluster found on chromosome 16 in all vertebrates (46). Hence, their transcriptional regulation and expression may be controlled by similar mechanisms, and these miRNAs may originate from the same cell type. According to our observations in NOD mice with new-onset diabetes, both miR-193b-3p and miR-365–3p are expressed by neutrophils and basophils. Remarkably, these miRNAs were more abundant in neutrophils and basophils isolated from the pancreas compared with those isolated from peripheral blood. Although the involvement of basophils remains largely understudied in type 1 diabetes, neutrophils have become increasingly recognized key players in the initiation and perpetuation of type 1 diabetes (reviewed by Battaglia (47), Huang et al. (48), and Giovenzana et al. (49)). Some studies described diminished peripheral-blood neutrophil counts coinciding with increased migration of these cells in the pancreas, in at-risk participants and those with type 1 diabetes (50,51), whereas other reports revealed increased, circulating platelet–neutrophil aggregates implicated in innate immune activation and migration, in at-risk children and children with new-onset type 1 diabetes (52). In the present study, broad transcriptomic and proteomic profiling using CITE sequencing of peripheral blood and pancreas-infiltrating immune cells of NOD mice with new-onset diabetes uncovered a mature (i.e., Egr1, Spi1), inflammatory (i.e., IL-1/IL-6, NF-κB family members, MAPK family members), and activated (i.e., Trem1, Gzmb, CD69) phenotype in pancreas-infiltrating neutrophils and basophils at disease onset compared with their circulating counterparts. Furthermore, STRING PPI networks revealed a close interaction between NF-κB, MAPK, and IL-1/IL-6 inflammatory pathways in pancreas-infiltrating granulocytes in NOD mice with new-onset diabetes. Interestingly, a previous study already reported that MAPK and NF-κB signaling pathways contributed to the regulation of miR-365–3p expression (53). Pretreatment with the MAPK/ERK inhibitor U0126 significantly reduced miR-365–3p expression, whereas forced overexpression of the canonical NF-κB family member p65 markedly increased miR-365–3p transcription. Moreover, ample evidence suggests that miR-365–3p is a direct regulator of IL-6 translation, because forced overexpression of miR-365–3p repressed IL-6 protein expression in a dose-dependent manner (53). On the basis of these observations, we speculate that pancreas-infiltrating basophils and neutrophils with a proinflammatory and activated phenotype most probably released miR-193b-3p and miR-365–3p into the circulation in response to MAPK or NF-κB pathway–activating stimuli in the inflammatory milieu of the pancreas. These granulocyte subsets may thus play critical roles in the subsequent response to current immunotherapies in people with type 1 diabetes, underscoring the importance of integrating the identified miRNA profile in future trial design.

Overall, we report in this study a reliable and noninvasive composite biomarker signature for predicting response to the L. lactis–based immunotherapy in new-onset type 1 diabetes. Apart from having a strong prognostic value, this biomarker signature also provides insights into the mechanisms of therapy nonresponsiveness and could be prioritized for individualized clinical decisions for individuals with newly diagnosed type 1 diabetes.

Acknowledgments. The authors thank Marijke Viaene and Jos Laureys for technical assistance, Jeroen Gilis for advice regarding statistical analysis, and Toon Ieven for assisting with blood collection. We also acknowledge Jana Roels, the VIB Single Cell Core Leuven, the VIB Flow Core Ghent, and the VIB Nucleomics Core Leuven for support regarding the single-cell technologies (vib.be/core-facilities). The authors thank the KU Leuven Flow and Mass Cytometry facility. The graphical abstract was created with BioRender.com.

Funding. This work was supported by grants from the Research Foundation Flanders (Fonds Wetenschappelijk Onderzoek [FWO] Vlaanderen, grant G.0C63.19N to C.M. and C.G.; FWO fellowship 1.1A02.20N to S.B.), the KU Leuven (C1/18/006), and by gifts for the Hippo & Friends type 1 diabetes and Carpe Diem funds for diabetes research.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors have contributed substantially to this study. G.Sa., G.L., G.V., C.G., and G.Se. designed the experiments. G.Sa., G.L., G.V., A.W., P.L., A.M., G.E.G., S.B., and G.Se. conducted the experiments and performed data analyses. G.Sa., R.S., N.V., and C.G. designed and carried out the bioinformatic analyses. D.E., S.C., and P.R. critically revised the article. G.Sa., C.G., and G.Se. wrote the manuscript. C.G. and G.Se. supervised the experimental design, data analysis and interpretation. C.M., F.D., C.G., and G.Se. revised the manuscript and supervised the study. The manuscript was approved for publication by all authors. G.Se. and C.G. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.