INOCA was defined as a positive (+) stress test (= ischemia), with no CAD at ICA (<20% stenosis). A positive stress test included abnormal functional studies including nuclear stress test, or stress echocardiography, with evidence of reversible ischemia on a nuclear study or reversible stress-induced wall motion abnormality on stress echocardiography, as determined by the attending cardiologist according to current ACC/AHA guidelines (8). The primary analysis compared the stress+ CAD- INOCA groups to control patients that were defined as stress- and CAD-. Invasive coronary angiography (ICA) was used to identify patients with obstructive CAD vs. those without CAD (Figure 1). Blood was collected in Tempus RNA stabilizing tubes prior to ICA, stored at −80°C and then profiled by RNAseq as described below. Coronary angiograms were digitally interpreted by the attending cardiologist. For statistical power, and a conservative definition of “no CAD”, initial analyses compared LOW CAD (<20% stenosis) to CAD combined (>20% stenosis) yielding groups of similar size.

Patients presenting for ICA related to non-emergent complaints of typical or atypical chest pain, exertional dyspnea, or other symptoms suggestive of CAD, provided written, informed consent for participation in this study under a protocol approved by The George Washington University IRB and the INOVA Fairfax IRB. Patients with heart failure, non-ST segment elevation myocardial infarction (MI) and ST elevation MI (STEMI) were excluded from the study. All patients presented in the period of October 2017 through September 2019, prior to the COVID19 pandemic. The design of the study is shown schematically in Figure 1. Patients admitted for ICA had three Tempus blood RNA tubes collected by peripheral venipuncture or an indwelling catheter just prior to ICA, which was typically up to 2 weeks after stress testing. After blood sampling, these studies were purely observational and did not alter the patient's clinical course. All relevant clinical data was captured for comparison to the transcriptome profiling.

Cardiac medical histories were evaluated by their attending cardiologists to determine CAD risk factors, according to the 2013 ACC/AHA Guidelines on the Assessment of Cardiovascular Risk (9). Hypertension was defined as a history of blood pressure ≥140/90 mmHg and/or treatment with anti-hypertensive medications. A family history of CAD was determined by MI or cardiac death in a first-degree relative. Diabetes mellitus was indicated by fasting glucose of ≥126 mg/dl and/or use of insulin or oral hypoglycemic agents. Current smoking was defined by active smoking within the most recent 3 months. Dyslipidemia was judged by the guidelines of National Cholesterol Education Program Adult Treatment Panel III or by treatment with lipid lowering medication.

As expected from patients undergoing elective ICA (n = 177), there were a significant percentage that did not have CAD > 20% (n = 79, 44.6%), which is consistent with nationwide rates for “no CAD” elective catheterizations (13). Of these 79 “no CADs”, there were more males overall (59.5%) than females, but were represented about equally in the stress+ and stress- groups (Table 1). The other demographic and lab values, such as age, race, BMI, blood pressure, diabetes, ejection fraction, and creatinine were not different between groups, except that stress-/CAD+ patients were significantly older and had higher ejection fractions. Likewise, the RNA yield (∼3.1 µg/3 ml blood) and the number of unique mapped reads (∼15 million reads per patient), were not different between groups.

The 79 patients without coronary stenosis >20% divided about equally into 40 patients that had a normal stress test, and 39 patients with an abnormal stress test indicative of ischemia, and thus INOCA. The RNAseq expression values were compared between groups using DeSeq2 to identify transcripts of potential interest. DeSeq2 identified 1,581 transcripts that were differentially expressed at an uncorrected p-value of 0.05 (Supplementary Data S1). This list was further filtered to include only transcripts that showed a greater than 1.5-fold increase or decrease in INOCA, yielding 66 transcripts increased, and 133 decreased (199 total, Supplementary Data S1). Only 2 transcripts, DAAM2 (down 4.9-fold) and LINC02056 (up 8.3-fold) were significant using the Bonferroni corrected p-value of 0.05, irrespective of fold-change (Figure 2). These 199 transcripts were used to understand the nature of the transcriptome changes in INOCA.

The primary list of 199 INOCA transcripts was analyzed by both manual and automated methods to understand the types of transcripts that were modulated. The transcripts were extensively analyzed by both automated (i.e., GeneCards, NIH DAVID, Ingenuity Pathways) and manual methods (i.e., PubMed) that considered: the direction of the change, the type of RNA involved (coding RNA/mRNA, microRNA, non-coding RNA, etc.), the cell types, functional pathways, annotations, and disease associations. Selected transcripts are shown in Tables 2, 3. Many of the miRNA and ncRNAs are poorly annotated and thus will require further study to understand their involvement. Focusing on the mRNAs, a clear pattern emerged: the upregulated transcripts fell into 2 groups, one associated with mucosal-associated invariant T (MAIT) cells and the other with B cells, especially memory B cells (Table 2). In many cases, the memory B cell transcripts could also be expressed by plasmacytoid dendritic cells (pcDC).

Conversely, the down-regulated transcripts were often strongly indicative of the granulocyte class of innate immune cells: basophils, eosinophils, and neutrophils (Table 3). Some of these mRNAs are considered consensus markers for neutrophils: CEACAM8 (aka CD66b) and CEACAM6 (CD66c), bacterial permeability increasing protein (BPI), CLEC4E, and TLR2. CEACAM8/CD66b, for instance, is a surface marker for neutrophils, similar to CD15, that is commonly used by our group, and many others, to identify and isolate neutrophils with classic polymorphonuclear phenotype from whole blood (14). A caveat is that some of these transcripts can also be expressed by monocyte-type cells, likely because they are involved in similar processes of innate immunity. A notable exception among the down-regulated transcripts is ZBTB16, which is considered a key marker of MAIT T cells but is decreased 1.57-fold in INOCA when most other MAIT-related transcripts were increased.

An in-depth examination of the 199 INOCA DEGs via both manual and curated methods indicates that at least 18 of the transcripts have known associations with disease, especially cardiovascular and autoimmune disease (Table 4). While of unknown function, C1orf105 has been associated with heart failure, atrial fibrillation (Afib), and CAD by genome-wide association studies (GWAS) (15). Recently, complement factor H (CFH), which inhibits the consumption of complement component 3 (C3), was found to modulate inflammation and the atherosclerotic process in a mouse model (16).

Notably, the most down-regulated transcript, dishevelled-associated activator of morphogenesis 2 (DAAM2), has variants linked to the development of nephrotic syndrome, in which the renal microvasculature is disrupted (17). Both DAAM1 and DAAM2 are required for myocardial maturation (18), and the genetic variants are known to substantially modify RNA expression levels (19).

Also noteworthy is the transcript for myeloperoxidase (MPO), which is a well-characterized neutrophil granule protein involved in NETosis (20). MPO is considered a “high-risk” biomarker for patients with acute coronary syndrome (ACS), CAD, heart failure, hypertension, and stroke (21). MPO inhibitors are in clinical development for several cardiac indications (22), including heart failure with preserved ejection fraction (HFwPEF) (23). The most increased transcript LINC02506, has been associated by bivariate GWAS with the connection between obesity and serum lipid levels (24).

Several genes had prior associations to CAD or MI: C1orf105, CHIT1, IL18RAP, PLIN4, TONSL-AS1, TPST1, VNN1, and WSCD2. Others were associated with autoimmune disease, such as systemic lupus erythematosus (SLE), ALS, and RA: CHIT1, IL18RAP, PHF24, and TPST1. In several cases, there was noteworthy overlap between the CAD/MI and autoimmune associations: CHIT1, IL18RAP, and TPST1. Erythroferrone (ERFE, aka CTRP15) protein levels were elevated in 190 CAD patients vs. 70 controls (25). Interestingly, serum ERFE levels associate with mortality and cardiovascular event in hemodialysis and CKD patients in 2 cohorts (26).

Pairwise correlations between all 199 INOCA transcripts, followed by hierarchical clustering, identified self-organizing clusters of transcripts that were interrelated (Figure 3). Initial analysis of these clusters identified patterns of their relationship to particular cell types, and key examples are highlighted for graphic clarity in Figure 4. A set of 3 clusters, including DKK2, SLC4A10, CFH, NTN4, and GLP1R, showed intercorrelations of up to 0.58, and were associated with MAIT cells.

A second tight cluster of transcripts, including GLDC, TNFRSF17, JCHAIN, IGLL5, and MZB1 showed intercorrelations of up to 0.82, and were strongly associated with memory B cells.

The largest cluster involved several well-known neutrophil-related transcripts, including a consensus marker for neutrophils, CEACAM8 (aka CD66b). Other well-known neutrophil transcripts, such as MMP8, TLR2, and CLEC4E were included in this cluster. Interestingly, the most strongly decreased transcript, DAAM2 was strongly intercorrelated in this cluster, with positive correlations to TLR2 (0.46), MAK (0.51), CLEC4E (0.66), and TPST1 (0.74). While leaning heavily toward neutrophils, many of these transcripts, such as TLR2, are also expressed, albeit at somewhat lower levels (∼50%) in monocytes, and thus clusters 4–6 are viewed as “neutrophil/monocyte” markers to indicate this uncertainty.

Thus, the transcripts organized themselves into groups that tracked the published associations of the transcripts or their proteins with particular cell types. While there were strong positive correlations within the clusters, there was a lesser, but consistently net negative correlation, up to −0.48 for CFH and DAAM2, between the MAIT cell cluster and the neutrophil cluster, as if there might be coordinated regulation (Figure 4). The cell types involved were pursued more directly with automated analysis using pre-curated data sets.

The cell types associated with the 199 INOCA DEGs was assessed in an unbiased manner by 2 different approaches. There are extensive, pre-curated datasets of gene expression on flow cytometrically purified cell types (B cells, T cells, etc.), that is integrated with single-cell RNA sequencing (scRNAseq) data sets. Likely the largest and most extensively curated dataset is the Human Protein Atlas (HPA), which has an extensive subset of data directed to RNA transcript levels in purified and scRNAseq analyzed human immune cells (12). This dataset reports the level of any transcript across essentially all known cell types, and conversely, identifies the top transcripts that are unique to each of those cell types. This allows 2 complementary ways to analyze the cell types involved in INOCA.

An unbiased analysis of the biological pathways related to the DEGs was conducted in Ingenuity Pathway Analysis (IPA), confirming that the major pathways affected by INOCA were related to the “neutrophil extracellular trap signaling pathway” (-log_4.3) and to a lesser, but still significant degree, 2 pathways related to rheumatoid arthritis signaling (Figure 6). The transcripts in this pathway are the family of transcripts reflected in Clusters 3–6 and having strong known relationships to neutrophil activation during innate immunity. The similarities with rheumatoid arthritis are intriguing, and will be discussed in depth later.

IPA further identifies a disproportionate number of the DEGs as being related to the interferon signaling pathway (Table 5). By this automated analysis, 24 of the 199 INOCA transcripts are known to be modulated by or involved in the interferon (IFN) signaling pathway. In addition to the IFN-regulated transcripts shown in Table 5, the full 1,581 DEG list (not restricted by fold change) contain numerous transcripts indicative of IFN activation, especially FASLG, ICOS, IFIT1B, IFITM2, IFNGR2, IFNLR1, and IKZF2, and IKZF3 (which is increased in INOCA but decreased in CAD). IRF2 and IRF2BPL, IFNGR2, IFNE were all decreased. STAT4, a key IFN signaling mediator, was increased. Thus, there is a strong signal for activation of the interferon pathway in INOCA, a phenomenon that has been extensively studied in relation to CAD [reviewed in (31)].

In the full cohort of 177 patients, we had previously employed 2 different RNAseq platforms to identify a pattern of changes associated with CAD that is best characterized as decreased expression of T regulatory (Treg)-related transcripts (27, 28). To determine whether CAD and INOCA had similar transcripts involved, we compared the two lists. There is not strong overlap (7.8%) of the exact transcripts of the current INOCA DEGs with our prior analysis of CAD transcripts.

Within the current dataset, we conducted a similar analysis, but used the subgroups that are free of confounding ICA+ and Stress+ patients (n = 60), thus comparing Stress-/ICA- to Stress-/ICA+, to identify transcripts associated with obstructive coronaries in non-ischemic patients. The transcripts associated with CAD in Stress- patients yielded 181 transcripts. Both the transcript overlaps on DEGs (Figure 7A), and the correlation of the levels of the transcripts in the groups were compared (Figure 7B). This narrower definition of CAD identifies 23 transcripts that are also INOCA DEGs. While only ∼10% overlap, the odds of such an overlap occurring by chance are extremely small (p = 4.6 × 10⁻²²). In all but 3 cases, the direction and magnitude of the change was similar in CAD and INOCA. The 3 exceptions, FN1, ZNF608, TMEM119, thus identify discriminant gene expression between CAD and INOCA.

Because transcripts could change in a similar manner in CAD and INOCA, but miss the DEG thresholds, the pooled list of CAD and INOCA DEGs was correlated between their expression when grouped by CAD vs. grouped by INOCA (Figure 7B). In this analysis, there is a net positive correlation of 0.44, which likewise indicates some shared properties of CAD and INOCA. It is striking that the LINC02506 transcript, which is the most elevated transcript in INOCA (up 8.3-fold), is also the most elevated transcript in this sub-analysis of CAD (up 8.7-fold). All possible correlations between the transcripts in the subgroups are shown in Figure 7B.

The recurring appearance of neutrophil-related genes suggested that INOCA could be related to some type of active infection. In prior publications, we identified neutrophil-related transcripts that respond to (1) free floating “planktonic” infections, such as pseudomonas in the lungs (32), (2) biofilm infections, such as those occurring in the appendix or on prosthetic joints (32), and (3) viral infections, such as SARS-CoV2 (33). Using these biomarkers, and other known transcripts such MPO and CTSG, we computed their expression levels (nRC) in the 4 major groups, and see no evidence for any type of acute infection in these patients. Although some patients had elevated values, there was no pattern by group (Table 6). The INOCA DEGs, CEACAM8, CTSG, and MPO are visibly lower in the INOCA group, but do not pass the non-parametric test used here, which differs from the DeSeq2 analysis that identifies DEGs. Thus, we can exclude an active infection as the cause of INOCA, but cannot exclude a prior infection that might have triggered a more stable immune defect.

The often-described increased prevalence of INOCA in females warranted analysis of INOCA transcripts as a function of sex. As shown in Figure 8, most transcripts do not differ as a function of sex, as exemplified by the reference example of actin B (ACTB). However, by contrast, the INOCA-related transcripts, MPO, CTSG, CEACAM8 and especially DAAM2, tended to be lower in females than males, and showed a greater change in the INOCA patients. Likewise, DAAM2 and CEACAM8 tended to be more elevated in patients over the age of 65 vs. under, but the difference was not statistically significant (not shown).

The increased expression of select genes related to B cell, pcDC, and MAIT cell activation, combined with decreased levels of neutrophil transcripts is an interesting and unique signature in blood RNA. The involvement of MAIT cells is highlighted by changes in several transcripts, including ZBTB16, aka PLZF, considered the master regulator of MAIT cells (34). It may be important that ZBTB16 transcript is decreased while almost all other MAIT associated transcripts are increased. MAIT cells have an intimate relationship to the gut and other microbiomes because they possess an invariant T Cell Receptor (TCR) that is tuned to bacterial metabolites, especially of B vitamins, that are presented on the MHC-related MR1 (35). MAIT cells would normally perform innate-like control of bacteria, but they are known to cross-stimulate other cell types such as B cells and conventional T cells in some autoimmune conditions (36). MAIT cells comprise about 5% of circulating CD3+ cells and have multiple influences on autoimmune diseases (34). For example, MAIT cells are chronically activated in patients with autoimmune liver disease (37). Interestingly, MAIT cells are known to be reactive against polyfluoroalkyl substances (PFAS), so called “forever chemicals” (38), and are functionally affected by smoking and COPD (39). Neutrophils can suppress MAIT cells, but conversely, activated MAIT cells can produce high levels of TNF-α that can induce neutrophil apoptosis (40), suggesting a generally antagonistic relationship between these cell types. This would be consistent with the net negative correlations in gene expression between the MAIT transcripts and neutrophil transcripts in Figure 4. The reduced neutrophil-associated transcripts that are seen in INOCA are not observed in the CAD patients (Table 3), and differs from published associations of NET-related activity with CAD (41). However, the relationship between NET activity and neutrophil RNA levels is quite complicated, but suggestive of the potential involvement of NETosis in INOCA, as discussed below.

The history of atherosclerosis research is punctuated with a variety of viral, bacterial, and autoimmune initiators for adverse vascular remodeling. While almost every virus and bacteria has been associated with atherosclerosis, recent studies on influenza and SARS-CoV2 (SCV2) lend very credible support to the likelihood of viral initiation of immune changes that lead to myocarditis and other cardiac diseases. Large studies have demonstrated a temporal association between influenza infection and myocardial infarction (42). Likewise, flu vaccines have a provable benefit in randomized control trials at preventing adverse events in high-risk CV patients (43).

The COVID-19 pandemic highlighted the connection between viral infections and adverse CV events. A signature event in SCV2 infection was an increased neutrophil to lymphocyte ratio (NLR) that led to striking NETosis and end organ damage in ICU patients (44). This was detectable in blood RNA as marked elevation of neutrophil transcripts, such as MPO, DEFA1, and neutrophil elastase (ELANE), but decreased levels of T cell transcripts (33). The data on SARS-CoV2 vaccines is more complex, with some evidence that RNA-based spike protein vaccines triggering myocarditis in some patients, but showing an overall protective effect on cardiovascular health in elderly “at risk” patients (45).

Recent sequencing of bacterial 16S rRNA in whole blood suggest that there are detectable changes in patients with acute coronary syndrome (ACS) (46). Changes in the gut microbiome have been associated with autoimmune inflammatory arthritis, and coincidentally affect B vitamin metabolism (47). Likewise, an increasing literature suggests that microbiome changes affect CAD progression via other metabolites such as imidazole propionate (48), TMAO, and short chain fatty acids such as propionate or butyrate [reviewed in (49)].

Regardless of the initiating event, the INOCA RNA profile includes many transcripts that would be potent immune modifiers. In a very striking coincidence, complement factor H (CFH, increased 1.5-fold in INOCA and 1.2–1.46 in CAD groups) was recently described as important to the atherosclerotic process in mouse models, and present in human atherosclerotic lesions (16).

Changes in the acquired (B cell and pcDC), invariant (MAIT T), and innate (neutrophil) immune system describe an immune dysregulation that parallels other known autoimmune syndromes such as lupus/SLE, systemic sclerosis, rheumatoid arthritis (RA), and others. Patients with SLE have more severe coronary microvascular disease (CMD) than symptomatic matched controls (50). Patients with RA have ∼1.5-fold increased risk of cardiovascular disease attributed to chronic inflammation (51). In women with CMD, about 10% have autoimmune rheumatic diseases (52). Further, disease-modifying anti-rheumatic drugs are showing beneficial effects on cardiovascular risk (53).

The pathway analysis shown in Figure 6 identified 3 pathways related to RA signaling that are also affected in INOCA. Those pathways involve 3 collagen genes (COL13A1, COL17A1, COL9A2), DKK2, FN1, interleukin receptors (IL17RB, IL18RAP), J chain, MMP8, SFRP1, and TLR2. The IL17 pathway is especially interesting therapeutically because it is a target of approved immunotherapies for psoriasis, which is an autoimmune disease with strong CV comorbidity (54). Two transcripts, JCHAIN (IgJ) and TNFRS17, both increased 1.7-fold and with high absolute expression level, have previously been identified as markers of plasmacytoid cells in autoimmune disease (55). Other data indicates that JCHAIN/IgJ is a marker of “plasmablasts” cells that are IgM and IgA secretors (56). A third gene, IGLL5 lambda chain, is a known marker of RA (57). IL18RAP is associated with a variety of autoimmune diseases including lupus, celiac, but especially with CAD via whole blood expression levels (58), and associated with MI via IL18 SNPs (59).

While similar to SLE, certain aspects of the INOCA pattern, however, are different. For instance, SLE has typically been thought to involve low density granulocytes (LDG) (60). However, RNA markers of LDGs, such as RSAD2, MX1, and IFI44 (60), are unchanged in INOCA. Rather, in INOCA the changes appear to be limited to more conventional granulocytes, especially the neutrophils. Antiphospholipid syndrome (APS) also exhibits cardiac manifestations consistent with microvascular disease and INOCA (61). However, the present data tends to discount APS because we previously saw strong HLA association in aspirin resistance (62), which is similar to APS, and yet we do not see HLA changes in the present studies.

It is accepted that innate immune cells, especially neutrophils, and unconventional T cells such as γδ T cells and MAIT cells are involved in microvascular remodeling, most notably in hypertension (63). Systemic sclerosis, which is caused by autoantibodies to anti-centromere proteins or anti-topoisomerase antibodies, among others, has well documented effects on the microvasculature (64). A systemic inflammation index, computed as the platelet count × neutrophil/lymphocyte ratio (NLR), is elevated in patients with microvascular dysfunction associated with cardiac syndrome X (CSX), which is ANOCA/INOCA-like (65). Recent pilot data suggests that NET markers in blood increase the predictive ability for all-cause mortality in patients with cardiac arrest (66).

Colchicine, a microtubule assembly inhibitor, in low doses has recently been approved to treat CAD, and its proposed mode of action is anti-inflammatory. Colchicine has known inhibitory effects on NET formation in the heart post-MI and in patients with ACS (67). Colchicine has a preferential effect on neutrophils because they generally lack the P-glycoprotein efflux pump, and thus inhibits intracellular mobilization and release of granule contents, especially elastases and defensins [reviewed in (68)].

A different perspective on the current data demonstrates that INOCA and CAD have both shared and unique RNA profiles. While only ∼10% of the transcripts are identical between the 2 disorders, the overall correlation of the transcripts between the two groupings of the dataset shows a net positive correlation of 0.45. An important consequence of this correlation is that prior analyses might have missed important changes in CAD. The presence of the INOCA patients in what was previously considered a “control” group for CAD, had the effect of masking some transcripts from being detected. For instance, LINC02506 is up 8.3-fold in INOCA, and up 8.7-fold in CAD, when the stress+ patients are omitted. However, in our prior combined analysis of the same patients that included stress+ patients and were grouped only by CAD, at the same 20% threshold, there is a much more modest 2-fold increase in LINC02506 levels. Thus, the presence of the INOCA patients in the “control” group of patients with stenosis <20% is a major confounding factor in identifying CAD-related gene expression. This is further complicated by the known potential for significant CAD of larger arteries but that is extra-luminal and thus non-obstructive. This will require a much more stringent definition of “controls” for future biomarker analysis of CAD.

The size of the study is quite powerful for a discovery-type study to generate hypotheses about the underlying etiology of INOCA. The analytical methods are diverse and provide a relatively consistent picture of activation of mem B, pcDC, and MAIT T cell transcripts, with decreased levels of neutrophil transcripts.

A significant limitation is that the clinical syndrome of INOCA is imperfectly measured by a stress test. In general, stress tests are prone to “false positive” results that can lead to unnecessary angiography, and thus, it is likely that some patients classified as INOCA in the present analysis are, in fact, not ischemic. There are a variety of more sophisticated tests, such as coronary flow reserve and acetylcholine provocation test during angiography, that could be used to measure microvascular function, but because they are not in routine use, this would require more costly invasive testing.

While the cohort is by far the largest RNAseq analysis of CAD and INOCA to date, there are intrinsic limitations to measuring 28 K transcripts in 177 subjects. Even a low noise level creates false positives, and negatives, that cannot be statistically corrected.

Another limitation is that while RNA can be quite accurately quantified by RNAseq, the RNA level has an unpredictable relationship to the encoded protein levels. In some cases, RNAs can be elevated because they are not being translated, and thus protein expression can be low. Conversely, it is known that many RNAs are degraded in the process of translation, and thus a reduced RNA level can be associated with elevated protein levels. Ideally, isolated neutrophils would be used for semi-quantitation of select protein targets to determine how the RNA levels related to protein levels. Further, functional analysis of NETosis and interferon pathways would be used to validate the RNA findings.

In retrospect, an in-depth cell type analysis by flow cytometry would have been helpful to evaluate the potential role of changes in the abundance of mem B cells, MAIT cells, and neutrophils, with only the latter being evident from a conventional laboratory blood count. However, this is quite difficult to accomplish in a busy cath lab setting. Furthermore, many prior studies have only documented changes in Treg cell abundance in CAD, though we are not aware of a similar analysis in INOCA.

The present results strongly indicate that future studies of CAD biomarkers in blood should carefully account for INOCA patients. The presence of small vessel disease will likely complicate any biomarker analysis.

An interesting question is whether the inflammatory transcript profile described herein would correct itself during treatment with colchicine, RA treatment or psoriasis immunotherapies. If so, the RNA profile could be used to track the efficacy and possibly dosing of these therapeutic modalities.