The Basic and Translational Science Year in Review: Confucius in the Era of Big Data

Abstract

Personalized medicine is an important goal for the treatment of rheumatic disease that seeks to improve outcomes by matching therapy more precisely with the underlying pathogenetic disturbances in the individual patient. Realization of this goal requires actionable biomarkers to identify these disturbances as well as pathways that can be targeted for novel therapy. Among advances in characterizing pathogenesis, Big Data provides an unprecedented picture of pathogenesis, with analysis of tissue lesions revealing disturbances that may not be apparent in blood. Big Data approaches include single cell RNAseq (scRNAseq) which can elucidate patterns of gene expression by individual cells. Galvanized by the Accelerating Medicine Partnership, a public-private initiative of the NIH, investigative teams have analyzed gene expression in cells in the synovium for rheumatoid arthritis and kidneys for systemic lupus erythematosus. A review of basic and translational research for 2018–2019 provides the progress in these areas. Thus, the studies on rheumatoid arthritis have identified subpopulations of immune cells and fibroblasts implicated in synovitis. For lupus, transcriptomic studies have provided evidence for widespread effects of type 1. Studies in progressive sclerosis have demonstrated changes associated with stem cell therapy as well as potential new targets for anti-fibrotic agents. Other studies using molecular approaches have defined new mechanisms for vasculitis as well as the potential role of the microbiome in inflammatory arthritis and systemic lupus erythematosus. Future studies with Big Data will incorporate the spatial relationships of cells in inflammation as well as changes of gene expression over time.

Introduction

In rheumatology, as in other medical specialties, two prominent lines of research are advancing rapidly along parallel tracks: personalized medicine and Big Data. Personalized medicine (also known as precision medicine) is based on the idea that clinical outcomes can improve by matching more precisely therapy with the underlying pathophysiology of disease in the individual patient (1). For personalized medicine to be successful, actionable biomarkers are necessary to select the right treatment among the many available options. In the absence of such markers, therapy remains empiric, leading to treatment delays as well as increased costs.

Progress in the development of new biomarkers for rheumatology has been comparatively slow despite extraordinary science on cellular and humoral abnormalities in rheumatic disease patients. Beyond tests for antibodies to citrullinated proteins (ACPA or anti-CCP), the actual markers used in practice have hardly changed in 50 years and, thus far, no surrogate markers have emerged. This situation reflects the limitations of older technology which can be said to represent Little Data. Little Data often comes in the form of a single data point (e.g., C-reactive protein, anti-DNA level, complement value).

In contrast to Little Data, Big Data is much more rich, detailed and encompassing, often multidimensional in nature. Genetics, epigenetics, genomics and other omics (e.g., proteomics, metabolomics) are all part of the Big Data revolution (2–5). Big Data also involve epidemiology as well as healthcare delivery (6). With Big Data technologies, the amount of data generated has increased by orders of magnitudes and it is not unusual for an experiment to produce millions and even billions of pieces of data. For Big Data, powerful computers are essential, with bioinformatics and novel statistical approaches providing the Rosetta Stone to translate the data into an interpretable form.

At its core, Big Data is a disruptive technology, necessitating new players, new skills and new ways of thinking. In Big Data, the important new players are the computational biologists who crunch numbers and develop algorithms to transform myriad data into accessible visual forms. The term disruptive is now popular in all parts of life, in most instances reflecting the impact of high speed computers on the organization and conduct of human activity. For medicine, the stakes of disruption are high since the investment in implementing Big Data into healthcare delivery is likely to be huge; this investment must yield better treatment outcomes to recapture the cost and to promote widespread implementation.

The Goals of Personalized Medicine

Many aspects of personalized medicine are old hat. Eyeglasses are a good example. While reading glasses that can be purchased cheaply in the supermarket are often satisfactory, they will be never as good (or personalized or precise) as those made by an optometrist. The optometrist can determine the abnormalities that impair vision and create a highly individualized pair of eyeglasses that can correct vision (near or far), adjusting for astigmatism as well. Contact lenses have the same properties. For the individual who wants the convenience of a single pair of glasses, bifocals and trifocals can readily be crafted.

In many respects, our approach to therapy of the rheumatic disease is similar to that of “dime store” reading glasses since therapy prescribed is without precise understanding of the actual abnormalities operative in the individual patient. While outcomes with biological and targeted synthetic DMARDs have improved impressively, the goal of personalized or precision medicine is to do even better. For conditions like rheumatoid arthritis (RA), systemic lupus erythematosus (SLE) and other inflammatory diseases, current biomarkers provide limited information. Big Data has the potential to change that.

Big Data

Consider the situation in genetics. While genetic testing long ago established the role of heredity in diseases such as RA, SLE and ankylosing spondylitis (AS), such testing never really became part of patient care; possible exceptions include HLA-B27 determinations in some cases of inflammatory back pain, especially in young boys and patients lacking radiographic findings. Similarly, even though the shared epitope is highly associated with RA, anti-citrullinated protein antibody (ACPA) testing provides similar information and is much easier to obtain. Rheumatoid factor (RF) testing often does as well (or even better) than ACPA testing despite its more non-specific nature.

With determination of single gene systems, the information is limited even for genes which have a significant impact on risk. The situation has changed dramatically because Big Data studies have identified a large number of genetic factors, with estimates of over 100 genes now found for SLE. The ability to characterize comprehensively the genetic profile of any individuals vastly increases risk prediction, even if patients with the same diagnosis differ in the number and array of genetic risk factors (4, 7, 8). The ability to use information from Big Data studies depends on the ease and cost of the studies to profile any individual. In this realm, enormous advances in genetic analysis now allows an essentially complete determination of an individual’s genome, including common and rare variants as revealed by exon sequencing. The costs of this analysis have dropped dramatically, placing this type of risk assessment in the realm of possibility.

Along the advances in genotypic analysis, functional and phenotypic analyses have benefited enormously from Big Data approaches including high dimensional flow cytometry and mass spectrometry to analyze cell populations. Mass spectrometry is related to flow cytometry but, instead of fluorescent reagents to identify proteins, antibodies are tagged with heavy metals to measure the protein content; mass spectrometry allows identification of the metal with less overlap than fluorescence markers and increases the number of markers that can be measured in the same experiment.

Among Big Data technology, single cell RNA-seq (scRNA-seq) can enumerate the transcriptional products of a single cell (2–5). In this technology, RNA is reverse transcribed into DNA which is amplified and sequenced using high throughput, next generation methods. From the DNA sequence, the transcribed genes can be inferred. scRNA-seq enormously enriches the picture of cellular activities that contribute to pathogenesis. Together, these Big Data technologies can identify and quantitate cell populations, seeking any perturbations associated with either a diagnostic category or disease state (e.g., activity). Disease can, therefore, be defined molecularly or functionally rather than clinically in terms of signs and symptoms.

The assay of cytokines provides a relevant example. While therapy for RA, AS, and psoriatic arthritis all involves monoclonal antibodies or other protein constructs (i.e., soluble receptors) to inhibit cytokine action, direct cytokine determinations are not part of the ordinary biomarker panels to evaluate disease clinically. Arguably, C-reactive protein (CRP) provides information on the amount of IL-6 in the system but the CRP marker is non-specific and shows elevation in RA where it can reflect the action of TNF-α. With the advance of Big Data, the field of biomarkers has entered a new era since techniques such as scRNA-seq can, in a single assay, vastly expand the number of cytokines determined along with the downstream genes activated. Furthermore, the assessment of cytokines can be precisely mapped to the producing cell while the genes activated downstream can be mapped to the responding cell.

These data can be incorporated into systems biology approaches to establish the interaction of genes and gene networks underlying pathogenesis in the individual. Thus, the potential for personalizing therapy is greatly enhanced since the landscape of pathogenic cells and their gene products can be assessed from peripheral blood; peripheral blood, the usual source of cells for such studies, however, is only part of the story.

In evaluating Big Data studies, a number of characteristics are important (2). These characteristics all begin with the letter V, although the number of these characteristics considered varies among sources. Table 1 lists these characteristics. The utilization of Big Data also requires consideration of many organizational and ethical issues (9).

Table 1:

The V’s of Big Data

Volume

Velocity

Variety

Variability

Veracity

Visualization

Value

The following are characteristics of Big Data which start with the letter V: volume, amount of data; velocity, the speed data are accessible; variety, types of data; variability, change in the data; veracity, accuracy of the data; visualization, presentation of the data to convey the meaning; and value, the utility of the data assembled.

The Chinese philosopher Confucius is famous for his many wise sayings. One that is relevant to the Big Data enterprise is the following: “Real knowledge is to know the extent of one’s ignorance.” While it is too early to determine whether Big Data equates to real knowledge, studies using these technologies have clearly demonstrated the extent of the ignorance on so many areas of immunology as applied to human disease. Indeed, prior studies using even the most advanced techniques have provided only limited information which is inherently superficial. The extent of ignorance is clear, with some of the excitement about Big Data the possibility that the field will gain enough new knowledge to make headway in many situations that previously were intractable.

Accelerating Medicine Partnerships

The last few years have been particularly important for the confluence of personalized medicine and Big Data. An important source of these data came from the Accelerating Medicines Partnership (AMP), a public-private initiative that combined resources of the NIH, industry and foundations (9, 10). The goal of the AMP has been to use new technologies, especially scRNA-seq, to define disturbances promoting the pathogenesis of RA and SLE at the molecular level. Importantly, the AMP applies new technologies to analyze events at the site of inflammation and injury, the synovium in RA and the kidney in SLE. In part, the goal of the AMP has been to identify new pathways and molecules as novel treatment targets, with an intense study of tissue lesions providing a new perspective on disease not possible from peripheral blood.

The years of 2018 and 2019 saw the publication of seminal papers from the AMP, although the application of Big Data is a worldwide enterprise that can involve individual laboratories as well as large consortia that include industry (11). Indeed, coincident with the growth of Big Data has been the expansion of so-called team science since Big Data studies necessitate big groups. The skill set for Big Data is large and diverse, with specialists in informatics joining forces with clinical, translational and basic investigators to conduct transformative research. The reach of these studies is far beyond the grasp of any one individual.

For the American College of Rheumatology (ACR) Annual Meeting held in Atlanta from November 8–13, 2019, I was asked to review the years in basic and translational research. This was a great honor as well as a great challenge since current research is voluminous and every Big Data paper is actually a tome. Each figure is in reality multiple figures since the average figure contains a large set of panels with its own letter; a stretch from letter “a” to letter “m” is not unusual. Furthermore, on-line supplements can accompany papers; these supplements often contain as much data as the regular paper, if not more. Finally, papers have accompanying data files which are deposited in large repositories so that other scientists and groups can also analyze them.

For this lecture, my review included publications from parts of both 2018 and 2019. Given this mountain of information, distilling tens to hundreds of hours of reading into 30 minutes required a drastic simplification and inevitable superficiality. Big Data is still in an early descriptive or hypothesis-generating stage to lay out the landscape. Later studies will fill in the details and link the observed patterns of gene expression into mechanisms. In this regard, a criticism of some thinkers or writers is that he/she cannot see the forest for the trees. In the next discussion, I will try to describe the forest since the Big Data experiments enumerate not only trees but leaves. A compendium of data on leaves, while very large, would probably not be very enlightening.

Similar to the lecture I delivered at the meeting, I will break down the discussion in terms of diseases and mechanisms. I will also include studies that consider papers other than those that represent Big Data; nevertheless, these studies benefit from the large amount of data available on disease mechanisms.

Diseases

Rheumatoid arthritis

While RA is a systemic disease, synovitis is its most characteristic feature and the major determinant of clinical outcome and patient quality of life. For many years, synovium was a major focus of investigation, although many of these studies involved tissue obtained at the time of joint replacement. As a result, the findings obtained may have reflected longstanding disease, including the effects of prior therapy and damage. Heterogeneity among patients as well as heterogeneity within different regions of tissue from the same patient also limited the information provided by biopsy. In practice, synovial biopsy as a source of biomarkers to stage disease or guide therapy was rarely if ever performed. Furthermore, the revolution in RA treatment which has been based on biologics and targeted synthetic disease modifying anti-rheumatic drugs (DMARDs) occurred in the absence of synovial biomarkers; the effects of some of these agents were so evident that biomarkers beyond the CRP and ESP were not needed to establish efficacy.

The rationale for a more intensive analysis of synovium to develop new treatment targets derives from the incomplete response of many patients to best therapy and the failure of current therapy to cure disease or at least induce longstanding remission, optimally drug free. According to this line of reasoning, analysis of immunological events in the synovium would reveal better targets of new therapy than possible with analysis of only peripheral blood. A corollary of this idea is that synovial biopsy at earlier stages of disease should be considered to identify relevant mechanisms that can be targeted.

A study by Zhang et al utilized Big Data approaches-sc RNA-seq, mass spectrometry, bulk RNA-seq and flow cytometry-to characterize T cells, B cells, monocytes and fibroblasts in a large number of RA patient samples obtained by ultrasound guided biopsies as well as at surgery (12). The studies demonstrated the presence of 18 unique cell populations in the synovium, allowing categorization of subpopulations that could not be distinguished on the basis of conventional immunohistopathology or flow cytometry. Importantly, using mass cytometry and transcriptomics, these studies showed the expansion of a number of cell types in the synovium. These cell types included sublining fibroblasts, pro-inflammatory monocytes, autoimmune-associated B cells and peripheral T cells and follicular T cell helpers. These studies also defined subpopulations of CD8 cells and mapped IL-6 production to fibroblasts and IL-1b to pro-inflammatory monocytes. In terms of the goal of finding new targets, certain chemokines (CXCL8, CXCL9 and CXCL13), cytokines (IFN-γ and IL-15) and surface receptors (PDGFRB and SLAMF7) all showed patterns of expression worthy of consideration.

By their nature, current technologies (e.g., scRNA-seq, multi-color flow cytometry and mass spectrometry) allow division of cell populations into an ever increasing number of subpopulations on the basis of the expression of a gene or gene clusters or the display of cell surface markers. Of cells involved in RA pathogenesis, monocytes/macrophages are very heterogeneous in function and phenotype and, therefore, a potential role in disease. In another study from the AMP using scRNA-seq, Kuo et al identified a subpopulation of pro-inflammatory macrophages defined as HBEGF+ on the basis of a cluster of genes that include HBEGF for heparin-binding EGF-like growth factor (13). This population is enriched in synovium and interacts with resident macrophages and cytokines, including TNF-α. These macrophages can increase the invasiveness of fibroblasts and, thereby, drive joint destruction.

In an extension of these studies, the investigators established in vitro synovial tissue cultures to elucidate the effects of medications on gene expression of the HBEGF+ macrophages that had been stimulated by fibroblasts and TNF-α. Perhaps not surprisingly, drugs to treat RA such as leflunomide, dexamethasone, naproxen and triple therapy all could block gene induction in this model. In contrast, methotrexate, a mainstay DMARD, was not active. These findings suggest that the HBEFG+ macrophages are only one of the drivers in pathogenesis, with methotrexate affecting other steps. In this regard, observations on the effects of anti-TNF on synovial cultures provided the impetus to develop TNF blockade as an RA therapy (14). Ex vivo models may, therefore, have a revival as the players are better identified and the effects of therapy can be analyzed more finely in terms of patterns of gene expression as opposed to the assay of a single analyte such as IL-1.

While the number of macrophages in the joint in RA can increase and drive inflammation, in the normal joint, these cells may have important physiological functions. In a study of cells in the mouse joint, Culemann et al demonstrated the presence of membrane-like structures that displayed what was dynamic and could provide a tight-junction barrier to protect the joint (15). As these studies showed, these macrophages may be replenished from a pool of locally proliferating monocytes. Thus, these studies extend the analysis of the role of macrophages to the normal joint and provide further evidence for the functional and phenotypic diversity of macrophages in the normal and autoimmune setting.

Systemic lupus erythematosus

As noted above, an important goal of the AMP has been to define cellular responses at the site of tissue inflammation and injury. In contrast to RA where a routine synovial biopsy is rare, renal biopsy in SLE is common and there is a wealth of information already available on the histopathology of SLE nephritis as well as certain aspects of gene expression. Thus, characterizing the landscape of lupus nephritis follows a long tradition of using tissue biopsy to stage disease and guide treatment. Notwithstanding the established use of biopsy, biopsy interpretation can be difficult and most relevant lesions for classification and decision-making remain uncertain, areas where Big Data could provide inroads.

Two studies from the AMP were published together in Nature Immunology and provided new information on the landscape of cells in lupus nephritis. For immune cells, Arazi et al used scRNA-seq to demonstrate the presence of 21 subsets of leukocytes in the kidney in active nephritis (16). These cells include multiple populations of myeloid cells, T cells, B cells and NK cells. Among these cells, B cells with the age-associated B cell phenotype were present as were monocytes in varying stages of differentiation. Other notable findings included the presence of an interferon response in most cells, the broad expression of the CXCR4 and CX3CR1 chemokine receptors and a correlation of responses of immune cells in the urine and kidney. The latter finding is especially notable since it would allow Big Data analysis at multiple times during the course of disease.

A related study by Der et al examined tubular cells in the kidney as well as keratinocytes in the skin by scRNA-seq (17). These studies demonstrated the presence of the type 1 interferon signature in both cell populations, with a high interferon signature as well as a fibrotic signature in tubular cells in the kidney associated with a failure to respond to treatment. Interestingly, the effects of interferon appear widespread when analyzed in terms of the genes that are induced. Taken in concert with the study of Arazi et al, these studies indicate that Big Data can provide novel information on a large number of cell populations present at sites of tissue inflammation, whether in the kidney or skin.

Even though conventional immunoassays indicated the presence of interferon in the blood of patients with SLE, the focus on this cytokine really followed studies by microarrays, an older technology for assessing gene expression. Despite the prominence of the signature in many patients, the relationship of the interferon gene signature (IGS) to disease activity has been uncertain. Catalina et al, therefore, used a bioinformatics approach to interrogate available microarray data sets to explore the relationship of the signature to disease activity (18). As this study showed, different IFN signatures are present in peripheral blood, with monocytes an important source of the expressed genes that comprise the signature. While the signature can vary over time, this study did not show a relationship to disease activity, likely because the gene signature is expressed by monocytes. Thus, the persistent expression of the IGS by monocytes, while pointing to an important immune disturbance in SLE, limits the utility of the IGS as a biomarker for activity as opposed to a target of therapy.

Genomics provides one approach for defining the functional capacity of a cell. Epigenetics provides another. To characterize the state of B cells in SLE, Scharer et al examined the DNA methylomes, chromatin accessibility profiles and transcriptomes of five B cell subsets in patients with SLE and healthy controls (19). The findings of this study are important in demonstrating that an SLE molecular signature is even present in resting naïve B cells, with enrichment of accessible chromatin in motifs for the AP-1 and EGR transcription factors.

Together, the Big Data studies on SLE provide an unprecedented picture of cells important in pathogenesis whether in the blood, kidney, skin or urine. These are cross-sectional studies, with longitudinal studies the next step to determine more fully changes that are associated with disease activity and treatment response. Future studies will involve three-dimensional reconstructions to indicate the spatial relationship of different cell populations.

Systemic sclerosis

A major advance in the treatment of systemic sclerosis (SSc) has been the use of myeloablation followed by hematopoeitc stem cells transplantation (HSCT) as shown in the SCOT (Scleroderma: Cyclophosphamide or Transplantation) trial (20). While the SCOT trial shows the benefit of HSCT, the impact on this modality on patterns of gene expression had not been investigated in detail, making the basis of the treatment response uncertain. As shown by Assassi and colleagues using microarray analysis, at baseline, blood of patients with SSc demonstrate notable findings in terms of modules in gene expression, with an increase of two IFN modules and one neutrophil model; this study also demonstrated increases in modules for cytotoxic/NK cells and erythrocytes.

Importantly, as demonstrated in longitudinal analyses by Assassi and colleagues, treatment with HSCT but not cyclophosphamide improved modules that had heightened gene expression (21). A proteomic study confirmed the improvement observed with HSCT. While gene expression analysis pointed to changes associated with the clinical improvement, this study used microarray to probe gene expression. Future studies using scRNA-seq and mass spectrometry to delineate more precisely changes in cell immune populations will undoubtedly clarify the nature of the therapeutic effects of HSCT and provide more actionable biomarkers.

Studies on transcriptional networks can also reveal new players in pathogenesis and, therefore, potential targets of new therapy. As shown by Wohlfahrt et al, the transcription factor PU.1 is an important regulator of pro-fibrotic gene expression by fibroblasts and can show upregulation in conditions associated with fibrosis (22). Importantly, studies in the mouse indicate that an agent called DB1986 can block PU.1 and can display anti-fibrotic effects in the bleomycin model. A related study by Soare et al analyzed the role of dipeptidyl-peptidase-4 in fibrosis (23). These investigators showed that the expression of DPP4 and the number of DPP4 positive fibroblasts is increased in the skin of patients with SSc and that in animal models overexpression of DPP4 can promote fibroblast activation. Furthermore, inhibition of DPP4 action by genetic knockout or pharmacologic inhibition can decrease fibrosis in both the bleomycin and graft-versus-host models. Inhibitors of DPP4 are called gliptins and are used in the treatment of diabetes because of their effects on incretin (GLP-1 and GIP) levels, raising the possibility of drug repurposing with an approved agent.

Mechanisms

Internal DNA sensing

One of the most exciting areas of immunology research relates to the recognition that host defense against bacteria, viruses and other organisms occurs inside of cells as well as outside. As now recognized, cells of the innate immune system have intracellular sensors to detect nucleic acids, both DNA and RNA, aberrantly present in the cytoplasm. These nucleic acids can arise from infection (i.e., bacteria or virus) as well as damage to mitochondria or even the nucleus during cell stress. These sensors include both toll-like as well as non-toll-like molecules. Mutations in these sensing systems as well as the DNase or RNase molecules that degrade cytoplasmic nucleic acids are associated with immune-mediated diseases that include increased interferon expression and clinical features reminiscent of lupus (24, 25).

Among these sensors, the cGAS-STING system has attracted great interest in diverse areas of pathology including host defense, autoimmunity, malignancy and senescence (26). Activation of this system begins with the interaction of DNA with a DNA sensor called cyclic guanosine monophosphate (GMP)-adenosine monophosphate (AMP) synthase (cGAS) to induce the synthesis of cyclic GMP-AMP (cGAMP), a novel dinucleotide mediator. cGAMP in turn binds to the protein STING (stimulator of interferon genes) which induces the robust stimulation of interferon and its downstream effects. Previous studies on the link between cGAS-STING and autoimmunity focused on the role of enzyme defects in the exonuclease TREX 1 in causing aberrant levels of cytoplasmic DNA and heightened interferon production in a condition known as Aicardi-Goutiere syndrome; studies on RNA reveal a similar phenomenon (24).

In an interesting study on a link between interferon and SLE, Gkirtzmanaki et al showed that one of the actions of interferon-α is to impair the degradation of mitochondrial DNA during autophagy (27). Autophagy is a process involving the delivery of cytoplasmic molecules to lysosomes for degradation and recycling to promote homeostasis. With defects in autophagy and intracellular clearance, levels of nucleic acids from damaged mitochondria can rise and trigger cGAS-STING activity. In SLE, high levels of interferon can impair this process, providing another way in which interferon can promote autoimmunity, further strengthening the rationale for targeting this molecule in SLE.

In SLE, stimulation of cGAS-STING and other internal nucleic acid receptors can result from cell damage which allows translocation of nuclear or mitochondrial DNA into the cytoplasm. Stimulation of these sensors may also occur with DNA or RNA that has entered cells of the innate immune system in the form of immune complexes, with the immune complex providing a transport vector to essentially transfect DNA into cells. In this model of pathogenesis, cGAS-STING represents an important target of therapy. A fascinating study by Dai et al demonstrated that acetylation of cGAS inhibits its activation and the generation of cGAMP (28). Remarkably, aspirin can cause this acetylation and block immune activation initiated by cGAS.

In other experiments, Dai et al demonstrated the effectiveness of aspirin using cells from a patient with Aicardi-Goutiere syndrome (28). To further show the relevance of aspirin as an inhibitor of autoimmunity, the investigative team demonstrated that aspirin can inhibit manifestation of autoimmunity in TREX1 knockout mice. Neither salicylic acid nor diclofenac, both NSAIDs that can block cyclooxygenase, were effective in this model. These studies demonstrate that the acetylation activity of aspirin is critical for the inhibition of cGAS.

Aspirin obviously has a long history in the treatment of rheumatic disease but its use has dwindled because of concerns of toxicity including gastrointestinal bleeding as well as salicilysm which can be fatal. Other NSAIDs have a better safety profile but even these drugs are used sparingly because of renal and cardiovascular effects. Long ago, clinicians felt that aspirin was an effective treatment of lupus especially at higher doses. The studies by Dai and colleagues raise the intriguing possibility that the efficacy of aspirin in SLE relates to its ability to block the cGAS-STING system. In this regard, high dose aspirin is able to inhibit the NF-κB pathway (29) suggesting that this venerable drug has three different anti-inflammatory actions related to inhibition of cyclooxygenase, NF-κB and cGAS-STING.

NETs

NETs or neutrophil extracellular traps are now recognized as an important element for host defense as well as tissue injury mediated by polymorphonuclear leukocytes (30). NETs are the product of a process called NETosis in which the nuclear membrane breaks down to allow mixing of chromatin with enyzmes in neutrophil granules. The resulting structure consists of strands of DNA that are decorated with granule contents, including anti-bacterial peptides, to create a mesh (or NET) that can entrap bacteria and other microorganisms and mediate their killing. While NETosis was originally conceptualized as a death process, viable cells can release NETs; the remaining cell, however, is devoid a nucleus and has limited capabilities. Since the DNA released during NET formation is high molecular weight, NETs heighten viscosity and confer on pus its thickness and gooiness.

A wide variety of stimuli can induce NET formation, leading to different types of structures including aggregates. While these structures can restrain and kill bacteria, they can also damage normal cells such as endothelium and contribute to vasculitis among other inflammatory conditions. A study on a genetically determined form of vasculitis called DADA2 provide a unique insight into the generation of NETs. DADA2 stands for deficiency of adenosine deaminase 2 and results from autosomal recessive loss-of-function mutations in the ADA2 gene. Clinically, DADA2 is characterized by vasculitis of medium and small vessels, livedo reticularis and stroke. Tissue lesions of DADA2 show abundant neutrophils and macrophages, with macrophages displaying a pro-inflammatory phenotype.

A study by Carmona-Rivera et al showed that patients with DADA2 have an increase in a neutrophil population known as low density granulocytes which are prone to NETosis (31). Importantly, adenosine can trigger NET formation by interaction with the A1 and A3 adenosine receptors in a process dependent on both reactive oxygen species (ROS) and peptidylarginine deiminase (PADI). As shown this study, in vitro NET formation can be blocked by the ADA2 enzyme, A1/A3 antagonists and an A2a receptor agonist. Interestingly, while NETs can mediate damage themselves, they may have other roles in the pathogenesis the data presented indicating that NETs can induce TNF-α production by macrophages. In addition to identifying potential therapies for this genetic form of vasculitis, these studies are important in highlighting a novel role of NETs in cytokine stimulation.

While NETs play a role in many inflammatory conditions, they nevertheless are structurally heterogeneous, with their properties varying with stimulus and disease. As shown by van Dam and colleagues, NETs in vasculitis differ from those in SLE (32). Whereas NETs in ANCA-associated vasculitis (AAV) display enrichment for citrullinated histones and depend on ROS and PADI, those in SLE contain HMGB1 and oxidized DNA; HMGB1 is a non-histone nuclear protein that is a prototype alarmin which becomes immunologically active when released from dead and dying cells. Of interest, as these studies showed, immune complexes in the blood of patients with SLE can stimulate NET formation whereas NET formation in samples from patients with AAV is independent of IgG. Along with showing the heterogeneity of NETs’ structure, findings in this paper are important in showing that ICs in SLE can drive NET formation, adding another mechanism by which ICs contribute to inflammation and damage.

Microbiome

The mammalian organism is a composite of genomes of mammalian origin as well as those of microbial origin including bacteria, viruses and fungi. Indeed, there are as many non-mammalian as mammalian cells in the average person, with microbes inhabiting multiple niches from the mouth to the skin to the gut. Together, these non-mammalian species comprise the microbiome. By far, the largest component of the microbiome exists in the intestinal tract or gut. While some of organisms can be cultured from the gut, the majority are recognized by sequencing of the ribosomal RNA, an enterprise made possible by rapid sequencing techniques of Big Data. The assemblage of sequence data into a catalogue of the organisms in the microbiome is another example of the wizardry of today’s computational biologists.

As shown primarily in studies of mice, the microbiome has profound effects on the host and vice versa (33–35). A genetic defect in an immune system gene can alter the composition of the microbiome and, correspondingly, the microbiome can impact immune system function. In mice, the effects of the host microbiome can be determined by manipulation of such germ-free derivation, antibiotics, and selective repopulation gut organisms. Diarrhea from Clostridium difficile is a classic example of a disease mediated by a change in the composition of the microbiome, with certain antibiotics setting up for the growth of C. difficile and the severe consequences.

A disturbance in the microbiome is termed dysbiosis and there has been keen interest in rheumatology as to the role of dysbiosis in the etiology of inflammatory diseases. In RA, for example, there is evidence for an increase in Prevotella species with active disease (33, 35). Genetics (e.g., the shared epitope) creates a strong predisposition to disease. Therefore, it is important to address whether any disturbance in the microbiome in RA results from the disease or from the genetic background.

A study by Asquith et al investigated the impact of HLA alleles on the gut microbiome in otherwise healthy individuals with HLA-DRB1 and HLA-B27, the former associated with RA and the latter with AS (36). The results of this study indicate that changes in the microbiome observed in patients with RA or AS may, in fact, represent the effect of the genetic background rather than the disease state itself; it is, of course, possible that the microbiome in turn shapes the immune system to make the emergence of a rheumatic disease more likely.

A study by Azzouz et al on the microbiome in SLE provides an intriguing picture on the role of the gut microbiome in the development of autoimmunity (37). As this study showed, patients with SLE have evidence of dysbiosis that is associated with disease activity. Furthermore, the study demonstrated that patients with SLE have impairment in the gut barrier that could lead to gut leakage and the exposure of the host to commensal bacteria. Other studies have provided evidence that bacteria may leave the gut and that extraluminal bacteria can elicit an immune response in a type of bacterermia (38).

Other findings in this study indicate an expansion of the Ruminococcous gnavus (RG) that was proportional to disease activity (37). Furthermore, evidence for a pathogenic link between dysbiosis and SLE is the finding of IgA-coated organisms in fecal samples and an increase in IgG to RG cell wall lipoglycan in patients with nephritis. Importantly, the study showed that patients from three different sites all had evidence of dysbiosis with increased RG, providing strong evidence for the relevance of these findings. As there are now ways to alter the gut microbiome (e.g., fecal), SLE may represent a setting in which this type of therapy can be contemplated.

Conclusion

Another Confucius saying relevant to the era of Big Data is, “Life is really simple but we insist on making it complicated.” While I doubt life is ever as simple as implied in this statement, the era of Big Data shows that life is really, really complicated and is likely to get even more complicated as analysis gets deeper and deeper. Thus far, most Big Data studies in rheumatology have been cross-sectional. Longitudinal studies that include the phenotypic and functional properties of multiple cell populations are the logical next step and will demand analytic strategies such as machine learning, artificial intelligence and neural networks. Cells operate in time and space. Three dimensions will be the next frontier of Big Data.

For the development of new treatments, some degree of simplification is essential, recognizing that medical practice cannot address all of the personalization that Big Data can provide. Drug development, clinicians and patients all insist on some simplification since contrary to Confucius, Big Data shows that life is really complicated and, despite revolutionary technology, ignorance will likely persist for many years. The good news is that we have enough knowledge to recognize that fact.

Figure 1.

Analysis of gene expression in synovium in rheumatoid arthritis. The figure illustrates the clustering of cells in rheumatoid synovium on the basis of the patterns of gene expression. For this purpose, a visualization technique known as a tSNE plot is used. tSNE stands for T-distributed Stochastic Neighbor Embedding and is an algorithm that allows presentation of large data sets. In this example, cell populations found in synovium are illustrated in a two-dimensional plot using color to distinguish the populations. Reproduced with permission from reference number 12 (Zhang et al).

Highlights.

• Big Data can provide biomarkers to elucidate pathogenesis and develop new therapy

• Multidimensional studies can define immune cell populations in tissue lesions

• DNA signaling provides internal defense against infection as well as trigger inflammation

• Genetic factors can influence the microbiome, with dysbiosis linked to autoreactivity

• Gene expression studies can identify responses to stem cell therapy in scleroderma