This review outlines the development of the Molecular Microscope Diagnostic System (MMDx) for heart transplant endomyocardial biopsies (EMBs) in the ongoing INTERHEART study (ClinicalTrials.gov NCT02670408). More details of MMDx are provided in other reviews,1,2 and other approaches to the molecular diagnosis of heart rejection have recently been reviewed.3 From the start of heart transplantation, there have been many sources of insight—function, survival, histology of EMBs, and recently an increasing number of biomarkers in body fluids. The INTERHEART study aimed to add to these dimensions by exploring the changes in gene expression in the EMB.
The scope of this review is as follows:
- The continuing importance of the EMB
- The molecular biology of rejection and injury states in organ transplants
- The strategies in the MMDx project and development of the EMB report
- Recent insights into subtle molecular gradients in EMBs
- The impact of injury on function and short-term graft loss
- The relationship between noninvasive biomarkers and MMDx
The Importance of the EMB
As acknowledged in recent reflections on the 50th anniversary of the first heart transplant,4-6 rejection remains an ongoing challenge. Additionally, we have become increasingly aware of the importance of parenchymal injury7: brain death, organ donation–preservation–implantation, and rejection are major “wounds” to the cardiac parenchyma, and wounds have long-term consequences. To understand the function and outcomes of transplanted hearts, we need to understand both rejection and injury, and the EMB remains a vital window on the disease states. New screening tests such as plasma donor-derived cell-free DNA (dd-cfDNA) have reduced pressure to perform biopsies, but the EMB is the cornerstone for exploring the biological mechanisms underlying rejection and parenchymal injury and their relationships to outcomes. Moreover, EMBs are essential for defining the relationship between noninvasive tests and the molecular processes in cardiac tissue.
Histologic Assessment of EMBs
Histologic assessment of EMBs has been crucial to the development of heart transplantation, using the International Society for Heart and Lung (ISHLT) guidelines first formulated in the 1980s in a working group led by Billingham8 and periodically updated.9,10 The ISHLT system inspired the development of the Banff criteria for kidney and other organ transplants.11 T cell–mediated rejection (TCMR) was identified as an interstitial mononuclear cell inflammatory process with parenchymal effects, including ultra-structural changes.12 Antibody-mediated rejection (AMR) was an early focus and continues to be a major issue.13-17 The principal histologic features of AMR were identified as intravascular activated mononuclear cells plus endothelial cells with large nuclei and expanded cytoplasmic projections.9 The ISHLT and Banff histology classification systems played a major role in the progress of organ transplantation (eg, as endpoints for clinical trials), have been mutually beneficial (eg, in understanding AMR),18,19 and have been critical to the development of MMDx.
General Challenges in Biopsy Assessment
All biopsy assessment, molecular or histologic, faces the challenges of sample adequacy: the potential influence of damage from previous biopsy procedures, multiple overlapping diseases, and intrinsic limitations of the diagnostic platforms (Table 1). Additionally, any process that assigns biopsies to a positive and negative group creates boundary errors: closely-related biopsies located near boundaries are placed into separate groups despite their similarity. The development of new biopsy assessment systems such as MMDx is complicated by the limits that institutional review boards (IRBs) place on the amount of tissue taken for research.
TABLE 1. -
General sources of error in biopsies
|Source of error
||How error is introduced
||Single small sample may not represent the organ
|Intrinsic variability within the organ: eg, kidney medulla versus cortex; atrophy-scarring; lung biopsies
||Nonspecificity of features (histologic or molecular), eg, arteritis can be AMR or TCMR
||Boundary error: near cutoffs
|Systems that rely on pattern recognition by individuals have intrinsic inter-individual variability
|Error within the rules
||Known limitations of the platform
|Weak phenotypes: “borderline,” “suspicious”
|Rare phenotypes: not present in the reference set
|Massive damage: beyond the range of the reference set
|AMR, antibody-mediated rejection; TCMR, T cell–mediated rejection.
Interobserver Variation in Pattern Recognition Systems
Medical diagnoses often rely on pattern recognition by single expert observers. This presents a challenge in reproducibility—variation or “noise” between observers assessing the same sample.20,21 This noise is not the fault of the experts, but rather an inherent feature of single-observer pattern recognition systems. Using the consensus of many independent assessments—an ensemble—reduces the noise, but is not generally practical in the clinic.
Interobserver variation in histologic interpretation of EMBs is apparent in the low kappa values for TCMR in heart transplants (0.28).22 Even in kidney transplant core biopsies, the agreement between 2 pathologists on whether the biopsy met the criteria for TCMR was only 50%.23 This must be remembered when we compare molecular findings to histology: molecular scores cannot agree with histology more than histology agrees with itself.
Issues Specific to Heart Transplantation
Heart transplants have 2 unique challenges: the relationship of the EMB finding to coronary artery changes, and the interpretation of the focal TCMR-like lesions called “quilty.”
Graft coronary artery disease—cardiac allograft vasculopathy (CAV)24—has been a focus in heart transplantation.25-31 Because EMBs cannot assess the coronaries, the relationship between EMB findings and CAV is unclear. Narrowing of the coronaries is often associated with parenchymal abnormalities, but whether these are caused by coronary artery abnormalities or simply associations reflecting shared injuries in the coronaries and myocardium remains unresolved. At least some CAV is associated with AMR as diagnosed in EMBs,32 as expected because AMR stresses all donor endothelium.
Heart transplants can have highly focal TCMR-like lesions in the endomyocardium, that is, “quilty” lesions.33 Quilty may appear focal, but these lesions may also be associated with diffuse changes as suggested by reports that quilty lesions are associated with inferior outcomes and micro-vasculopathy.34,35 Quilty lesions were a concern in the MMDx project36 because IRBs initially limited the INTERHEART study to 1 piece per biopsy. MMDx now recommends including at least 2 pieces.
THE MMDx PROJECT
The aims of molecular analysis of biopsies are as follows:
- To discover the disease states and their mechanisms in a continual loop of discovery and application, learning from the new biopsies.
- To recalibrate the existing histology diagnostic system.
- To develop diagnostic services for clinical biopsies to make discoveries accessible to clinicians.
- To define relationships between gene expression in the organ and noninvasive assessments such as dd-cfDNA.
Molecular biopsy assessment has been heavily used in cancer, often as extensions of histology diagnoses. Many cancers are characterized molecularly to subclassify histologic diagnoses, offer prognostic information, guide treatment, and identify potential targets of treatment (such as HER2).
As illustrated in Figure 1, molecular phenotyping of biopsies does not aim to agree with conventional assessments. All are estimates of the true disease state, which is never completely known. The goal of molecular analysis is to define abnormal molecular states and describe their relationship to histology, function, noninvasive tests, donor-specific antibody (DSA), and outcomes.
MMDx Biopsy Processing
Sample processing for MMDx is visualized in Figure 2. The tissue collected for MMDx is immediately immersed in RNAlater, which kills the RNAses that otherwise would rapidly degrade the mRNA. The biopsy in RNAlater is stable at ambient temperature for approximately 7 to 10 d, and indefinitely if frozen. Snap-freezing the biopsy without RNAlater could be used to stabilize the biopsy, but frozen tissue is harder to handle and ship, and can be rapidly destroyed if mishandled. MMDx can be performed on formalin-fixed paraffin-embedded (FFPE) tissues but has not yet been operationalized because formalin irreversibly compromises the RNA.
RNA is isolated from the biopsy tissue, labeled, and hybridized into the PrimeView microarray. The microarray has a silicon chip with about 500 000 features, each printed with a specific oligonucleotide that will hybridize with complementary labeled sequence fragments in the biopsy mRNA (Figure 3). The intensity of the signal indicates the degree of hybridization. Several features are combined as a probe set to reflect the expression of a specific gene. The result is expression levels for 49 495 probe sets representing 19,462 genes in the genome, expressed as a “CEL file.” This underestimates the huge complexity in the mRNAs from the genome, for example, alternative splicing, multiple promoters, and complex reading frames but is more than sufficient as a genome-wide representation. The CEL file is then processed using MMDx software to generate an automatic report listing the principal molecular scores for that biopsy. EMB processing takes 48 h from the receipt of the samples, longer than for kidneys (24 h) because the small EMB yields less total RNA (average 1.4 μg from EMBs versus 3.0 μg from 3 to 5 mm sections of kidney biopsy cores) and therefore requires more time for labeling the mRNA.
As noted above, the MMDx project operates as a continual discovery–application–discovery loop. Ongoing clinical trials provide new biopsies to update the reference set and re-derive the machine learning algorithms periodically. This requires that the discovery and application technology be identical so that the system can compare each new biopsy to the reference set. MMDx can read single biopsies in real-time and does not require waiting for large batches.
For this reason, microarrays are currently the technology platform of choice. They are an older but thoroughly standardized technology with guaranteed availability and high manufacturing standards. The use of microarrays permits us to discover molecular processes, create machine learning algorithms, and interpret new clinical biopsies by comparing them to the locked reference set. We periodically add new cases to the reference set and re-derive the algorithms, and are currently refining the machine learning algorithms with a reference set of 3000 biopsies.
RNA sequencing is superior to microarrays as a discovery system but is batch-dependent with potential batch-to-batch variation and is not economical for a single biopsy. However, we are monitoring sequencing options as technology evolves. MMDx has retained the RNA isolated from each reference set biopsy to permit switching entirely to an RNA sequencing platform if this can be shown to be feasible, cost-effective, and superior.
The Machine Learning Strategy for MMDx Algorithms
The probe set measurements obtained from the microarray CEL file are interpreted as gene expression using Bioconductor37 software. MMDx-Heart software developed in the ATAGC translates the gene expression measurements into probabilities of specific disease states using ensembles of machine learning-derived algorithms.
Machine learning algorithms for assessing rejection and injury have been developed using both “supervised” and “unsupervised” approaches. Supervised algorithms were trained by comparing a positive to a negative class, using labels previously assigned to the biopsies, for example, those that distinguish TCMR from “everything else.” Unsupervised methods look for clusters or groups within the data but are not provided with any biopsy labels. We compare these groups with the conventional labels for the biopsies afterward to interpret the assigned clusters. Unsupervised methods include principal component analysis (PCA) and archetypal analysis (AA), a form of clustering.
Interpretation of the MMDx findings draws on knowledge from experimental results, previous human biopsies, and the molecular biology literature. For example, molecular analysis of mouse transplant isograft models (i.e., with genetic identity between donor and recipient) identified transcripts induced by transplantation injury, the cardiac injury transcript set or “cIRITs.”36 The cIRITs have been useful in exploring the molecular phenotypes of injury. Because injury responses are shared across organs, we also used mouse kidney isografts to study injury,38 and mouse allografts to study TCMR, which is superimposed on other injuries.39 The injury-induced transcript changes in allografts are at first identical to those in isografts but are massively amplified when TCMR begins to add new injury.40 Thus, TCMR not only produces a characteristic inflammatory process but also parenchymal injury that has similarities with injury induced by the transplantation process.
The Need for Precise Central Measurements
We define precision as the reproducibility of the readings when the same isolated RNA is run on 2 microarray chips, assuming the same reagents and device are used. The precision of mRNA measurements on the PrimeView microarrays is >99%.41-44 Like many new molecular tests (eg, dd-cfDNA measurements), MMDx is performed centrally rather than as kits for local use. This requires shipment of the biopsy, but avoids center-to-center and machine-to-machine variation in measurements. Variation between local measurements on different machines is a well-known problem when technology is offered as kits to local laboratories, for example, in DSA measurements.45,46 Clinical application of MMDx requires that we setup one or more MMDx service laboratories. Each MMDx service laboratory must be a duplicate of the discovery laboratory, using identical technology, reagents, and software, and calibrating its output against the central laboratory regularly to avoid “drift.”
Precise quantitative measurements are essential when using machine learning algorithms, which analyze complex relationships such as ratios among the transcript measurements expressed as continuous numbers. The molecules associated with rejection and injury overlap, and several processes are often operating simultaneously. Distinct processes share features: for example, TCMR and AMR show increased expression of interferon gamma (IFNG)–inducible genes, but injury also manifests an increase in expression of IFNG-inducible genes, usually at a lower level. Machine learning can distinguish specific disease processes even when they share features.
Machine Learning Can Identify Errors
Machine learning analysis of the molecular changes associated with disease states as defined by histology labels has the potential to identify errors in histologic diagnoses (“labeling errors”). The algorithms identify the molecules associated with the labeled biopsies (eg, AMR), but can also use the molecular expression to identify biopsies that have been mislabeled. For example, in the MMDx-Kidney project, we initially compared the biopsies called AMR with those called non-AMR using the contemporary Banff definition, which relied heavily on C4d staining. The machine learning software discovered the molecular features of AMR and pointed out that many C4d-negative biopsies not called AMR had the diagnostic AMR-associated molecular changes.47,48 This is how we discovered the extent of C4d-negative AMR47 and then of DSA-negative AMR.49 Similarly, as outlined below, molecular classifiers trained on histology diagnoses agreed more with the molecular diagnoses than with the histology labels on which they were trained.50
Annotating the Top Genes “Associated With” Versus “Selective for” a Disease State
A disease manifests 2 types of changes: disease-associated (“abnormal”) and disease-selective (distinguishing different types of abnormal). Diagnostic systems use both, but disease classification depends on the latter. We must be precise in framing the question when discussing the molecules affected by a disease. For example, identifying all genes that change expression in AMR (i.e. associated with AMR) requires algorithms that compare AMR with normal tissue. Identifying genes that selectively change expression in AMR versus TCMR or injury requires algorithms that compare AMR with everything else, including TCMR and injury.
MOLECULAR BIOLOGY OF THE REJECTION AND INJURY STATES
The molecular features of rejection have been explored in human transplant biopsies, informed by mouse studies where possible. An illustration of immunologic processes in TCMR and AMR is presented in Figure 41 (with permission).
Alloantigen triggers clones of naive and memory antigen-specific (“cognate”) T cells to become activated effector T cells. This includes memory T cells capable of recognizing alloantigen51 because of cross-reactions between previous viral antigens and major histocompatibility complex (MHC) that is heterologous immunity.52 Naive T cells are triggered by antigen on antigen-presenting cells (APCs) in secondary lymphoid organs, with strict costimulation requirements; memory T cells have less strict requirements.
In TCMR, cognate effector T cells home to the graft, recognize antigen and cross the microcirculation, and engage APC. This activates both the cognate effector T cell and the APC, and the resulting mediators recruit inflammatory cells to create a highly structured interstitial inflammatory reaction—the TCMR landscape.53 The local parenchyma deteriorates, dedifferentiates, and manifests an injury response. Whether parenchymal deterioration reflects the actions of cytotoxic enzymes such as perforin and granzymes is unclear: knockout mouse studies suggest it does not.54 As seen in immunosuppressed transplant patients, TCMR can be focal, patchy, or diffuse, perhaps reflecting the number of cognate effector T cells. In hearts, the effects of a very limited number of cognate effector T cells may explain quilty lesions.
The transcripts most selective for heart TCMR are those expressed by activated effector T cells, for example, CD2, CD3D, and genes for the T cell alpha/beta receptor chains TRBC1 and TRBC1.53 Of interest, inhibitory checkpoint molecules such as CTLA4 are also expressed in TCMR, indicating that an element of the regulation is operating in the tissues during TCMR episodes.
The findings in INTERHEART show that TCMR is time-limited, becoming rare after 5–10 y.7 We believe that this is probably due at least in part to exhaustion of the cognate T cell clones in the secondary lymphoid organs due to immune checkpoints because checkpoint inhibitors can trigger severe TCMR.55
We distinguish 3 AMR scenarios in transplant biopsies:
- Hyperacute AMR: devastating injury, caused when an organ is transplanted into a patient with high levels of preformed donor-specific antibodies. This is essentially never seen in contemporary transplantation because of meticulous crossmatching.
- Type 1 AMR: early-onset AMR due to a memory response in sensitized patients. This is usually prevented by crossmatching, and seen primarily in cases where a weak DSA exists but the clinical risk is deemed acceptable.
- Type 2 AMR: later onset due mainly to de novo DSA production. This is by far the most common group. Overall, type 1 has better outcomes than type 2.56
AMR is diagnosed histologically by microcirculation inflammation, and molecularly by 3 types of AMR-selective transcripts: natural killer (NK), IFNG-inducible, and endothelial. The transcripts most selective for AMR are those in activated NK cells (eg, GZMB, KLRD1, and CCL4), endothelial cells (eg, ROBO4), and IFNG-induced genes (eg, PLA1A and CXCL11).18,19
AMR is usually mediated by DSA against HLAs, capable of activating AMR effector mechanisms (eg, NK cells). However, as discussed later, the DSA may not be detectable. DSA is polyclonal and heterogeneous. The DSA capable of triggering AMR may be only a fraction of the circulating DSA. To trigger effector mechanisms, IgG must not only be able of binding antigens but also of forming multimers such as hexamers, as shown for complement binding.57 However, AMR is unlikely to be mediated by complement because complement inhibitors do little to arrest the progression of AMR, except perhaps in very early Type 1 AMR.58
We conceptualize that DSA binding to donor MHC molecules on endothelium has little direct effect on the donor microcirculation unless the DSA IgG molecules are “AMR-effector-competent” and form specialized multimers to induce AMR effector systems. Based on the transcripts selective for AMR, the key AMR effector mechanism is NK cell activation by their CD16a Fc receptors binding IgG-HLA multimers on the donor endothelium, possibly potentiated by “missing self” recognition.59,60 NK cell transcripts such as granulysin (GNLY) and granzyme B (GZMB) are highly associated with AMR. The microcirculation adapts to this ongoing stress by increased expression of endothelial transcripts such as ROBO4 and loss of endothelial transcripts such as ECM1. As detailed below, molecularly defined AMR is often relatively well-tolerated compared with the parenchymal injury caused by TCMR, but in the long term has serious deleterious effects.
The science of AMR is evolving rapidly with the recognition that AMR is often DSA-negative, and that minor AMR-related changes can exist in many biopsies currently called “no rejection,” as discussed below. Recently, Coutance et al61 found significant relationships between microvascular inflammation (MVI) and expression of AMR-related transcripts, frequency and mean fluorescence intensity of DSA, and acute graft dysfunction, but continuing analysis of the relationships between MVI, molecular changes, DSA, function, and above all effects of intervention are needed.
Parenchymal Injury and Response-to-Wounding
Injury is universal among all organ transplants: the stress of donation, preservation, and implantation “wounds” the graft tissue and evokes a response-to-wounding that is programmed in all tissue. The response-to-wounding in parenchyma, matrix, and microcirculation has several components:
- Increased expression of injury molecules.
- Dedifferentiation of the parenchyma: loss of the molecules that maintain normal function, for example, solute carriers.
- Recruitment of innate immunity, for example, macrophages and polymorphs.
- Stabilization of irreversibly injured sites by atrophy-fibrosis.
Mapping molecular injury changes—response-to-wounding—is critical because of the following:
- Injury-induced changes are strong molecular correlates of dysfunction and graft loss.62
- Injury-induced innate immunity shares features with adaptive immunity (rejection), creating diagnostic challenges.
- Rejection causes injury that will persist after rejection is treated or subsides.
- Routine histology does not assess parenchymal injury as well as molecular measurements.38
THE DEVELOPMENT OF THE MMDx-HEART SYSTEM FOR CLASSIFYING REJECTION
The MMDx-Heart rejection system uses the expression of kidney-derived rejection-associated transcripts (RATs) to characterize rejection and distinguish TCMR from AMR. The system emerged over many years in the INTERHEART study, a large international collaboration.63,7,19,36,50,64
The main effect of immunosuppressive agents on the EMB seems to be mediated exclusively by the degree to which they control TCMR and AMR. We have not seen any effect of specific immunosuppressive agents on the details of the gene expression in the biopsy, in heart or kidney transplant biopsies, even in mouse studies (unpublished observations).
The First 3-Archetype Model in 331 EMBs
We serially developed the MMDx algorithms as the INTERHEART reference set accumulated, first in 331 EMBs, then in 889 EMBs, and recently in 1320 EMBs.64
We created the first archetype model using 331 EMB microarray results. Because of the published low kappa values for EMB histology,22 we relied primarily on unsupervised methods (PCA and AA). The kidney-derived RATs distributed heart transplant EMBs and kidney transplant biopsies similarly in PCA, supporting the use of RATs for analyzing EMBs.
AA based on the expression of RATs was used to assign archetypal groups within the population of EMBs. AA identifies “archetypes,” theoretical biopsies representing the most typical molecular features of a group. Every real biopsy is assigned archetype scores reflecting its “closeness” to each of 3 archetypes, and membership to one of the archetype groups based on its highest archetype score.
This initial 3-archetype model (3AA) is shown in Figure 5A.64 The biopsies (the reference set) are distributed in PCA by their expression of RATs: principal component 1 (PC1), no rejection versus rejection; principal component 2 (PC2), AMR (up) versus TCMR (down). The reference set biopsies are colored by their molecular rejection archetype group (blue = AMR, red = TCMR, and gray = no rejection).
The histology diagnoses for the biopsies are shown in Figure 5B.64 Despite many discrepancies, the archetype assignments largely correlated with the histology diagnoses. The agreement of molecular scores with histology was less than in kidney transplant biopsies, as expected given that EMBs are more difficult to assess histologically than kidney core biopsies and have more interobserver disagreement. The AMR archetype scores also correlated with DSA. The landscape of gene expression in heart AMR19 was similar to that in kidney AMR.18
The genes correlating with the TCMR score were associated with activated effector T cells, for example, TRBC1, CD3D, and CD2. The transcripts correlating with the AMR score were NK-expressed (eg, GZMB), IFNG-induced (eg, PLA1A and CXCL11), and endothelial (eg, ROBO4). The “No rejection” score correlated with high expression of normal heart transcripts (“HTs”) and low expression of rejection transcripts.
The 4 Archetype Model in 889 EMB Identified Biopsies That Were Inflamed Due to Recent Injury
Upon collection of 558 new biopsies, we reassessed how the machine learning algorithms derived in the original 331 biopsy cohort performed. We confirmed that algorithms derived in 1 cohort were stable in the new cohort of 558 biopsies.7
Using the combined cohort of 889 biopsies, we addressed the question of whether the inflammation in hearts with injury from the transplantation process might be simulating TCMR. We developed a 4 archetype model that recognized 4 groups (Figure 5C7): 3 of these 4 groups were similar to the 3AA No rejection (NR), TCMR, and AMR with a fourth identifying inflamed biopsies with no rejection but extensive inflammation due to injury (“NR-Early injury”). These separated from the other biopsies in principal component 3 (PC3) (Figure 5D7, PC2 versus PC3). These biopsies were infiltrated by myeloid inflammatory cells such as macrophages and had high expression of some RATs because innate immunity/inflammation shares genes with adaptive immunity/rejection.
The genes most strongly correlated with high NR-early injury scores included many expressed in macrophages such as FCGR2A, C3AR1, CD14, and CD163. Transcripts expressed in healthy hearts had reduced expression in injured hearts, indicating dedifferentiation. The early inflamed biopsies were often called TCMR by histology.7 Thus‚ expression of RATs allowed us to recognize not only rejection-related inflammation but also inflammation (innate immunity) related to injury and response-to-wounding.
The intensity of inflammation in NR-Early injury biopsies reminds us of the dynamic relationship between cardiac myocytes and macrophages in health and disease.65-74 Cardiac myocytes are notoriously vulnerable to injury and can evoke inflammation. This may involve vesicles released by myocytes reminiscent of neural exophers75 that contain cardiac mitochondria and their DNA, which can trigger innate immunity.
Developing Supervised Classifiers
To develop ensembles of estimates designed to improve the accuracy of our diagnoses, we trained binary rejection classifiers using the gene expression data associated with either histologic or molecular diagnoses.50 The molecular classifiers trained on molecular diagnoses correlated with both the molecular and the histologic diagnoses, molecular better than histologic. Classifiers using molecular measurements and trained on histology labels predicted the molecular diagnoses better than the histologic labels on which they had been trained (areas under the curve [AUCs] >0.87 versus <0.78). The MMDx diagnoses (discussed in detail below) showed highly significant agreement with histology (P < 0.001), albeit with discrepancies.
An automated random forest score using all molecular scores closely predicted expert MMDx diagnoses, confirming the potential for fully automated signouts.50 Nevertheless, trained observers do continue to sign out each biopsy to comment on challenges such as boundary issues or mixed phenotypes.
THE CURRENT MMDx REPORT
The current MMDx report for heart transplant biopsies (Figure 650) uses ensembles of diagnostic algorithms50 and is licensed for commercial service. The main elements of the MMDx-Heart report are numbered 1–6 in Figure 650:
- The 2-panel figure showing the new biopsy (triangle) in relation to the reference set of 889 existing biopsies. This expresses the uniqueness of each new biopsy rather than simply assigning it to a class, and shows its closest neighbors in the reference set. This reference set is colored by the 4-archetype (4AA) model,7 with the biopsies distributed in PCA based on their expression of RATs and colored by their rejection archetypes (gray = NR, blue = AMR, red = TCMR, and cyan = recent injury). In PCA, RAT expression distributed the reference set with the more homogenous normal samples concentrated towards the left, and the more heterogenous abnormal biopsies spread out towards the right. The cloud appears triangular in PC2 versus PC1, with normals to the left. Rotating the cloud 90° (PC2 versus PC3) makes it appear more circular, with normals in the center, a projection to the right that represents recent injury (PC3), and rejection deviating to the left.
- The automatically-assigned archetype scores in the 2 published models, one for 3 archetypes and one for 4 archetypes, including recent injury.
- The scores for molecular classifiers trained on molecular diagnoses. These classifiers were developed to predict whether this biopsy has no rejection, AMR, or TCMR.
- Expression values for relevant gene sets and single genes. These molecular measurements are associated with AMR, TCMR, recent injury, or normal parenchyma.
- Space for users to add additional notes. Such data are never used for report generation.
- The interpretation of rejection and injury by a trained expert (usually an experienced technologist). Complex or ambiguous results are reviewed centrally.
NEW DEVELOPMENTS IN RAT-BASED ANALYSIS: “MINOR” CHANGES, GRADIENTS, AND DSA-NEGATIVE AMR
We updated the analysis in 1320 EMBs,2 which classified 853 biopsies as NR. To explore heterogeneity within NR, we developed a 5-archetype RAT model that used a “Minor” archetype score to subclassify the 853 NR biopsies as 462 “Normal” (NR-Normal) and 359 “Minor” rejection changes (NR-Minor), while retaining the small group of 32 “Early injury” biopsies (NR-Early injury).
Compared to Normal biopsies, the Minor biopsies (NR-Minor) had a mild increase in AMR transcripts and an increased probability of being DSA-positive. These biopsies were often assessed by histology as “TCMR1,” indicating that histology also recognized subtle abnormalities.63 This Minor group in heart transplants is similar to a group of kidney transplant biopsies with a subtle “minor” increase in AMR transcripts and increased DSA, despite being called “No rejection” by MMDx and histology.76
Figure 763 shows 1320 biopsies emphasize the gradients in RAT expression. Biopsies with a low probability of rejection (“No rejection”) are heterogeneous: they can be “Normal,” “Minor,” or “Early injury.” Biopsies with high AMR-related transcript expression can be “possible AMR” or “AMR.” Biopsies with high TCMR-related transcript expression can be possible TCMR or TCMR (Figure 7A63). Mixed rejection biopsies express both AMR- and TCMR-associated transcripts. In Figure 7B,63 rotating the data cloud again reveals PC3 (early injury).
Complexity of Associations With DSA
In MMDx, only 66% of hearts with molecular AMR are DSA-positive, similar to observations in kidney transplants.77 In 1320 EMBs,63 DSA positivity increased in a gradient: NR-Normal 24%; NR-Minor 34%; possible AMR 42%; AMR 66%. This paralleled the gradient of increasing expression of AMR-related transcripts: NR-Normal to NR-Minor to possible AMR to AMR. Thus rejection is not adequately described as binary classes such as “AMR” and “No rejection.” Molecular AMR-associated changes (“AMR-ness”), the probability of DSA, and TCMR-associated changes (“TCMR-ness”) are all gradients.
Although molecular disease-associated changes are gradients, clinicians must still make binary decisions such as treatment versus no treatment, considering the unique position of the biopsy plus prior probabilities, previous history, recent treatment, infections, suspected nonadherence, and the risks of harm from treatment.
ASSOCIATIONS OF RAT AA SCORES WITH SHORT-TERM GRAFT LOSS
INTERHEART is a cross-sectional study, and as such has limited long-term follow-up. Nevertheless, we can study the hearts that progressed to failure after the biopsy, designated as “x” in Figure 8.63 There was no strong relationship between rejection in the biopsy and subsequent graft loss: many losses occurred after biopsies that did not show molecular rejection. Unexpectedly, the biopsies with active AMR (blue symbols) virtually never progressed to failure in the follow-up period, probably because active AMR produces little short-term parenchymal damage. We also see this in early-stage AMR in kidney transplants.78
Although a population basis, there was a paucity of short-term failures in the hearts assigned a molecular AMR diagnosis. However, within individual cases there may be a more aggressive course and more parenchymal injury. To date, observations have shown that molecular AMR cases have done well in the short term but appear to be associated with deterioration and Late-injury changes, which are inherently deleterious in the long term.
NEW CLASSIFICATION OF PARENCHYMAL INJURY BASED ON INJURY-ASSOCIATED TRANSCRIPT SETS
We are developing a new molecular classification of parenchymal injury and atrophy-fibrosis in the cardiac parenchyma, independent of the RATs and the rejection-related states.62 This allows us to assign each EMB its rejection (RAT)-related phenotype and its parenchymal injury state.
In the 1320 EMBs from 645 patients previously studied for rejection,63 we measured the expression of 10 injury-associated transcript sets: 5 induced by recent injury; 2 reflecting macrophage infiltration; 2 normal heart transcript sets with reduced expression in injury (dedifferentiation); and immunoglobulin transcripts, which correlate with time and atrophy-fibrosis. Expression of injury transcript sets correlated with impaired function—a gradient in left ventricular ejection fraction (LVEF) (Table 262). The injury-increased sets were highest and normal heart transcript sets were lowest when LVEF was <30, establishing the relevance of these measurements.
TABLE 2. -
Expression of injury-related pathogenesis-based transcript setsa,b
in hearts with LVEF > 45, 30 ≤ LVEF ≤ 45, and LVEF < 30
||Injury-related transcript set
||LVEF > 45 (N = 789)
||30 ≤ LVEF ≤ 45 (N = 51)
||LVEF < 30 (N = 19)
||Type I ANOVA LVEF interval P value
|Expressed in macrophages
||9 × 10−09
||6 × 10−10
|Increased in recently injured hearts
||5 × 10−08
|Kidney transplant derived injury-induced transcript sets
||2 × 10−04
||3 × 10−05
|Highly expressed in normal heart
||5 × 10−12
||2 × 10−15
|Increased in time
||4 × 10−16
bThe transcript sets were empirically derived in human cell lines, human transplants, and mouse models. They reflect biological processes relevant to rejection and injury.
AMAT, alternative macrophage-associated transcripts; ANOVA, analysis of variance; cIRIT, cardiac injury-repair induced transcripts; DAMP, damage-associated molecular pattern transcripts; HT1, heart transcripts set 1; HT2, heart transcripts 2; IGT, immunoglobulin transcripts; IRITD3, injury repair induced transcripts day 3; IRITD5, injury-repair induced transcripts day 5; IRRAT, acute kidney injury transcripts; LVEF, left ventricular ejection fraction; QCMAT, quantitative constitutive macrophage-associated transcripts.
PCA of injury transcript set measurements separated 2 dimensions of variance: PC1, increasing injury; PC2, increasing atrophy-fibrosis (Figure 962). Archetypal clustering assigned each biopsy to 1 of 5 injury groups: 376 No-injury; 526 Mild-injury; 110 Moderate-injury; 87 Severe-injury; and 221 Late-injury.
The injury groups differed in mean days posttransplant, particularly moderate to mild to no injury, reflecting the recovery of donation-implantation injury (Table 362). There were gradients in the degrees of abnormality in the injury transcript sets across the groups. Severe-injury had extensive loss of normal transcripts (dedifferentiation) and an increase in macrophage and injury-induced transcripts, whereas Late-injury was characterized by high immunoglobulin transcript expression, similar to atrophy-fibrosis in kidney transplants (Table 362). Many late heart transplants —nearly a fifth of all INTERHEART biopsies—manifest a late-injury phenotype with expression of the transcripts that in kidneys are associated with atrophy-fibrosis.79,80
Table 3. -
Mean time posttransplant and transcript set scores in the five parenchymal injury states
Injury variables assessed
Parenchymal injury states
Mean days post-transplant (median)
Mean transcript set scores in each parenchymal injury state
|Expressed in macrophages
|Increased in recently injured hearts
|Other injury-induced gene sets that correlate with cIRITs
|Highly expressed in normal heart
|Increased in late transplants
Bolding marks the highest expression groups in each row; italicized indicates the lowest.
A These were the 10 transcript sets used in the principal component and archetypal analyses. We also included the MCAT and BAT because they are also increased in atrophy-fibrosis, but they were not included in the PCA or AA.
Figure 1062 shows the importance of injury states in predicting graft loss. In all biopsies, Severe-injury and Late-injury increased the risk of short-term graft failure (Figure 10A62). The Moderate-injury group had few failures, presumably because these were the earliest biopsies. Severe-injury and Late-injury significantly increased graft loss even in hearts with no rejection (Figure 10B62). Within TCMR (Figure 10C62) and within AMR (Figure 10D62), the Severe- and Late-injury states play a major role in determining the risk of short-term graft failure.
Relating Injury to Rejection
Because this parenchymal injury analysis is independent of the RAT-based analysis, we were able to address a question that has long been of interest: how much injury does rejection induce? We compared parenchymal injury states to the rejection states assigned by RATs (Table 462). Biopsies with TCMR almost always had parenchymal injury: 85% had Severe- or Late-injury. In contrast, biopsies with early AMR had little injury beyond that expected in all early biopsies, and Severe-injury was uncommon in AMR. Late AMR biopsies were often assigned the Late-injury state. This is consistent with the natural history of AMR in kidney transplant results, where early-stage AMR initially spares the parenchyma and has little risk of 3-y graft loss but progresses toward a late-stage AMR with atrophy-fibrosis and increased graft loss.76,78,81
TABLE 4. -
Parenchymal injury states in hearts with TCMR and AMR
Modified rejection signouts
||Parenchymal injury states
|No-injury (N = 376)
||Mild-injury (N = 526)
||Moderate-injury (N = 110)
||Severe-injury (N = 87)
||Late-injury (N = 221)
|TCMR-related, including mixed and pTCMR (N = 127)
||Number of biopsies
Mean days post-transplant
|AMR-related including pAMR (excluding mixed) (N = 340)
||Number of biopsies
Mean days post-transplant
a Principal injury groups in each RAT signout group (> 15% of row total) are bolded.
AMAT, alternative macrophage-associated transcripts; AMR, antibody-mediated rejection; cIRIT, cardiac injury-repair induced transcripts; DAMP, damage-associated molecular pattern transcripts; HT1, heart transcripts set 1; HT2, heart transcripts 2; IGT, immunoglobulin transcripts; IRITD3, injury-repair induced transcripts day 3; IRITD5, injury-repair induced transcripts day 5; IRRAT, acute kidney injury transcripts; pAMR, possible AMR; pTCMR, possible TCMR; QCMAT, quantitative constitutive macrophage-associated transcripts; AMR, antibody-mediated rejection; TCMR, T cell–mediated rejection.
Table 562 shows the importance of Severe- and Late-injury for determining LVEF within the hearts with rejection. In hearts with TCMR, those with Severe-injury or Late-injury had the lowest LVEF, and all failures were in these groups. In hearts with AMR, those with Severe- or Late-injury had the lowest LVEF and 14 of the 20 failures, 12 in hearts with Late-injury.
TABLE 5. -
Relating parenchymal injury states to LVEF and graft loss in AMR and TCMR
|Modified rejection sign-outs
||Injury archetype group assignment
||N per group
||Mean time posttransplant
||Median time posttransplant
||Number of graft failures within 3 y
|TCMR + pTCMR + Mixed
(N = 127)
AMR + pAMRb (N = 340)
Severe- and Late-injury rows are bolded to highlight these data.
a Within TCMR biopsies, a t-test comparing LVEF in biopsies with Late-injury + Severe-injury versus those with No-injury + Mild-injury + Moderate-injury was significantly different‚ P = 0.007.
b Within AMR biopsies, a t-test comparing LVEF in biopsies with Late-injury + Severe-injury versus those with No-injury + Mild-injury + Moderate-injury was significantly different‚ P = 1.43 × 10−7.
AMR, antibody-mediated rejection; AMAT, alternative macrophage-associated transcripts; cIRIT, cardiac injury-repair induced transcripts; DAMP, damage-associated molecular pattern transcripts; HT1, heart transcripts set 1; HT2, heart transcripts 2; IGT, immunoglobulin transcripts; IRITD3, injury-repair induced transcripts day 3; IRITD5, injury-repair induced transcripts day 5; IRRAT, acute kidney injury transcripts; LVEF, left ventricular ejection fraction; pAMR, possible AMR; pTCMR, possible TCMR; QCMAT, quantitative constitutive macrophage-associated transcripts; TCMR, T cell–mediated rejection.
Emerging Concepts: The Effects of TCMR, AMR, and Parenchymal Injury
- The interstitial TCMR process has extensive effects on the adjacent parenchyma, triggering a response-to-wounding that involves expression of injury molecules, dedifferentiation (loss of function), and ultimately atrophy-fibrosis. TCMR-induced injury can persist after TCMR activity has been suppressed by treatment.
- AMR is microcirculation stress with a spectrum of intensity and stage that initially has relatively little effect on the parenchyma. AMR may be well-tolerated for months or years but can lead to late atrophy fibrosis. (Note that severe early type I AMR can trigger ischemia and severe parenchymal injury.) The prevalence of AMR is greater than previously appreciated, often DSA-negative, and includes “minor AMR-like changes.”
- The state of the parenchyma determines function and prognosis via the cumulative burden of injury. Many hearts with damaged parenchyma do not have active rejection but may reflect the impact of previous rejection that has been suppressed by treatment or spontaneously subsided at the time that the injury state is diagnosed.
- The coronary arteries and the parenchyma share many stresses, including TCMR, AMR, and other injuries (in addition to their unique stresses). CAV and the Late-injury parenchymal phenotypes may identify the same hearts. A detailed comparison of coronary imaging at the time of biopsy and the molecular injury and rejection states in the biopsy is needed.
APPLICATIONS AND CONSIDERATIONS OF DIAGNOSTIC PLATFORMS
The MMDx project has focused on the discovery, aiming to understand the genome-wide molecular changes in heart transplants and suggest the mechanisms that mediate these changes. To impact care, this knowledge must be made available to clinicians.
Central Genome-wide Measurements
Thermo Fisher has licensed MMDx and uses a centralized MMDx molecular platform to perform the assays in the Kashi service laboratory (https://kashilab.com/). MMDx uses intact ribonucleic acid (RNA, not FFPE) in a highly standardized and precise measurement platform, with the resulting expression data interpreted through machine learning algorithms. The central approach is necessary to ensure that the MMDx machine learning algorithms are stable and to avoid the problem of inter-laboratory variability. The disadvantage is the time needed for the shipment of specimens.
Local Laboratory Testing
The lessons from the MMDx project can be applied in local laboratories using kits and platforms such as quantitative RT-PCR, RNA sequencing, NanoString using the B-HOT panel of 770 transcripts,82 reverse transcriptase multiplex ligation-dependent probe amplification,83 and others. Some of these have been developed to use FFPE samples rather than taking additional EMB pieces.
The relative diagnostic performance and cost-effectiveness of different approaches should be supplemented by estimates of the cost-benefit relationships, for example, costs in terms of labor, reagents, capital, transportation, overhead, versus benefits in terms of improved guidance for management, including the use of expensive and potentially dangerous treatments. Ultimately, accurate diagnosis is still central to management, but diagnostic tests cannot change outcomes unless they lead to better management decisions.
RELATIONSHIP OF MMDx TO NONINVASIVE BIOMARKERS
Many noninvasive biomarkers are in development for monitoring heart transplants and avoiding unnecessary biopsies. Plasma dd-cfDNA measurements are one example.84-86 We are studying the relationship between dd-cfDNA and MMDx findings in transplanted hearts (Trifecta-Heart, ClinicalTrials.gov #NCT04707872) and kidneys (Trifecta-Kidney, #ClinicalTrials.gov NCT04239703). The Trifecta-Heart studies are in progress, but the Trifecta-Kidney results are currently available and are instructive with regard to what we expect for hearts.87 Trifecta-Kidney87 compared the fraction dd-cfDNA (%dd-cfDNA) taken just before the kidney transplant indication biopsies to the molecular phenotype of the biopsy. In the first 300 biopsies, the %dd-cfDNA was highest in active AMR, active TCMR, and mixed rejection but was often low in late-stage AMR and in some biopsies with TCMR. Parenchymal injury also caused some elevation of %dd-cfDNA. The %dd-cfDNA correlated more strongly with molecular variables than histologic variables.87,88
CONSIDERATIONS SUPPORTING CONFIDENCE IN MMDx ASSESSMENTS
MMDx is intended to extend knowledge, not compete with histology. No test has absolute truth, but we have reasons for believing that MMDx is more likely to be correct than histology when MMDx and histology are discrepant89 (Table 6 derived from Reeve et al89). MMDx assesses many additional features, measured with high precision; correlates better with outcomes; uses continuous scores rather than semi-quantitative or binary scores; has trained on histology features and thus incorporates the lessons of histology; is supported by experimental studies; has guided updates in the histology classifications (eg, C4d-negative AMR); assesses parenchymal injury38 more reliably than histology; and correlates better with %dd-cfDNA measurements.87,88
TABLE 6. -
Reasons why we believe that MMDx is more likely to be correct than histology89
|MMDx assesses many more features selected from genome-wide microarray measurements, measured objectively with high precision (reproducibility).
|In predicting a phenotype with a well-defined gold standard such as survival, MMDx outperforms histology.
|MMDx outputs are continuous rather than semiquantitative or binary, which is especially particularly important when a result is near the boundary.
|MMDx has trained some classifiers on histology features and therefore incorporates the lessons learned from histology in the reference set.
|Extensive molecular studies in laboratory animals support the interpretation of the transcript changes used in MMDx.
|MMDx findings have been used to update the Banff classification: recognition of C4d-negative AMR and DSA-negative AMR.
|MMDx can assess recent injury and dedifferentiation,
which cannot be estimated in routine histology.
|Donor-derived cell-free DNA measurements—a valid objective estimate of the probability of rejection—has been shown to correlate better with MMDx than with histology.
AMR, antibody-mediated rejection; DSA, donor-specific antibody; MMDx, Molecular Microscope Diagnostic System; TCMR, T cell–mediated rejection.
The key teaching points from MMDx-Heart studies are summarized in Table 7.
TABLE 7. -
Teaching points from the MMDx analyses
||TCMR is a highly damaging process, inducing severe parenchymal injury and late atrophy-fibrosis. TCMR-induced injury will likely persist after TCMR has been treated.
||AMR is usually a microcirculation stress but is not highly damaging to the parenchyma and is well tolerated for months or years. But AMR led to late atrophy-fibrosis, which will persist after the AMR activity has subsided or is suppressed by treatment.
||AMR scenarios should be distinguished as:a. Type 1 AMR: early-onset AMR due to a memory response in sensitized patients. This is usually prevented by crossmatching unless it is decided to proceed despite increased risk.b. Type 2 AMR: later onset due mainly to de novo DSA production. This is by far the most common group. Overall, type 1 is better tolerated than type 2.58
||Parenchymal injury induced has 3 aspects:a. Response of the parenchyma, matrix, and microcirculation: expression of injury molecules.b. Dedifferentiation of the parenchyma: loss of the molecules that maintain normal function.c. Recruitment of inflammatory cells, eg, macrophages and polymorphs.
||There is a subtle “Minor” state of AMR-induced molecular change in biopsies currently not recognized as AMR, with slightly increased AMR transcripts, increased DSA, and often called TCMR1 by histology.
||DSA-positivity occurs in a gradient: AMR 66% > possible AMR 43% > No rejection-Minor 36% >No rejection-Normal 24%.a. 34% of diagnosed AMR is DSA-negative.63b. We hypothesize that the cause of DSA-negative AMR changes in undetected donor-specific alloantibody, usually HLA-specific, perhaps amplified by missing self NK recognition.
AMR, antibody-mediated rejection; MMDx, Molecular Microscope Diagnostic System; TCMR, T cell-mediated rejection.
NEW DIRECTIONS FOR MMDx-HEART
As the INTERHEART (ClinicalTrials.gov #NCT02670408) and Trifecta-Heart (ClinicalTrials.gov #NCT04707872) studies continue, we aim to define the stages of AMR and TCMR and further explore parenchymal injury; get longer follow-up and more outcomes, and address the relationship of MMDx to CAV, DSA, biomarkers, function, and outcomes. The mission of our molecular analysis of EMBs is to maximize the utility of the EMB, which corresponds nicely with the parallel efforts to reduce the number of unnecessary EMBs. Our current priorities include expanding the reference set, currently at 3200 EMBs. Above all, we aim to establish the effect of interventions, using MMDx and other readouts. The goal is not simply to develop molecular insights, but to use the insights to change care.
1. Halloran PF, Famulski KS, Reeve J. Molecular assessment of disease states in kidney transplant biopsy samples. Nat Rev Nephrol. 2016;12:534–548.
2. Halloran PF, Madill-Thomsen KS; INTERLIVER study group. The molecular Microscope® diagnostic system meets eminence-based medicine: a clinician’s perspective. Am J Transplant. 2020;20:2964–2965.
3. Benck L, Sato T, Kobashigawa J. Molecular diagnosis of rejection in heart transplantation. Circ J. 2022;86:1061–1067.
4. Koomalsingh K, Kobashigawa JA. The future of cardiac transplantation. Ann Cardiothorac Surg. 2018;7:135–142.
5. Mehra MR. Heart transplantation at 50. Lancet. 2017;390:e43–e45.
6. Stehlik J, Kobashigawa J, Hunt SA, et al. Honoring 50 years of clinical heart transplantation in circulation: in-depth state-of-the-art review. Circulation. 2018;137:71–87.
7. Halloran PF, Reeve J, Aliabadi AZ, et al. Exploring the cardiac response to injury in heart transplant biopsies. JCI Insight. 2018;3:123674.
8. Billingham ME, Cary NR, Hammond ME, et al. A working formulation for the standardization of nomenclature in the diagnosis of heart and lung rejection: Heart Rejection Study Group. The International Society for Heart Transplantation. J Heart Transplant. 1990;9:587–593.
9. Berry GJ, Burke MM, Andersen C, et al. The 2013 International Society for Heart and Lung Transplantation Working Formulation for the standardization of nomenclature in the pathologic diagnosis of antibody-mediated rejection in heart transplantation. J Heart Lung Transplant. 2013;32:1147–1162.
10. Billingham M, Kobashigawa JA. The revised ISHLT heart biopsy grading scale. J Heart Lung Transplant. 2005;24:1709.
11. Solez K, Axelsen RA, Benediktsson H, et al. International standardization of criteria for the histologic diagnosis of renal allograft rejection: the Banff working classification of kidney transplant pathology. Kidney Int. 1993;44:411–422.
12. Hammond EH, Yowell RL. Ultrastructural findings in cardiac transplant recipients. Ultrastruct Pathol. 1994;18:213–220.
13. Berry GJ, Angelini A, Burke MM, et al. The ISHLT working formulation for pathologic diagnosis of antibody-mediated rejection in heart transplantation: evolution and current status (2005-2011). J Heart Lung Transplant. 2011;30:601–611.
14. Szymańska S, Grajkowska W, Sobieszczańska-Małek M, et al. Reclassification of C4d-Positive Endomyocardial Biopsy (EMB) according to new International Society for Heart and Lung transplantation (ISHLT) 2013 categories for reporting pathologic antibody-mediated rejection (pAMR): preliminary data from a polish single-center study. Ann Transplant. 2015;20:351–356.
15. Hammond MEH, Revelo MP, Miller DV, et al. ISHLT pathology antibody mediated rejection score correlates with increased risk of cardiovascular mortality: a retrospective validation analysis. J Heart Lung Transplant. 2016;35:320–325.
16. Kobashigawa J, Colvin M, Potena L, et al. The management of antibodies in heart transplantation: an ISHLT consensus document. J Heart Lung Transplant. 2018;37:537–547.
17. Fedrigo M, Leone O, Burke MM, et al. Inflammatory cell burden and phenotype in endomyocardial biopsies with antibody-mediated rejection (AMR): a multicenter pilot study from the AECVP. Am J Transplant. 2015;15:526–534.
18. Venner JM, Hidalgo LG, Famulski KS, et al. The molecular landscape of antibody-mediated kidney transplant rejection: evidence for NK involvement through CD16a Fc receptors. Am J Transplant. 2015;15:1336–1348.
19. Loupy A, Duong Van Huyen JP, Hidalgo L, et al. Gene expression profiling for the identification and classification of antibody-mediated heart rejection. Circulation. 2017;135:917–935.
20. Silver N. The Signal and the Noise: Why So Many Predictions Fail–but Some Don’t. 1st ed. The Penguin Press; 2012.
21. Topol E. Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. 1st ed. Basic Books; 2019.
22. Crespo-Leiro MG, Zuckermann A, Bara C, et al. Concordance among pathologists in the second cardiac allograft rejection gene expression observational study (CARGO II). Transplantation. 2012;94:1172–1177.
23. Reeve J, Sellarés J, Mengel M, et al. Molecular diagnosis of T cell-mediated rejection in human kidney transplant biopsies. Am J Transplant. 2013;13:645–655.
24. Mehra MR, Crespo-Leiro MG, Dipchand A, et al. International Society for Heart and Lung transplantation working formulation of a standardized nomenclature for cardiac allograft vasculopathy—2010. J Heart Lung Transplant. 2010;29:717–727.
25. Loupy A, Coutance G, Bonnet G, et al. Identification and characterization of trajectories of cardiac allograft vasculopathy after heart transplantation: a population-based study. Circulation. 2020;141:1954–1967.
26. Hammond EH, Ensley RD, Yowell RL, et al. Vascular rejection of human cardiac allografts and the role of humoral immunity in chronic allograft rejection. Transplant Proc. 1991;23:26–30.
27. Hammond EH, Yowell RL, Nunoda S, et al. Vascular (humoral) rejection in heart transplantation: pathologic observations and clinical implications. J Heart Transplant. 1989;8:430–443.
28. Prada-Delgado O, Estévez-Loureiro R, Paniagua-Martín MJ, et al. Prevalence and prognostic value of cardiac allograft vasculopathy 1 year after heart transplantation according to the ISHLT recommended nomenclature. J Heart Lung Transplant. 2012;31:332–333.
29. Daud A, Xu D, Revelo MP, et al. Microvascular loss and diastolic dysfunction in severe symptomatic cardiac allograft vasculopathy. Circ Heart Fail. 2018;11:e004759.
30. Cheng R, Kransdorf EP, Wei J, et al. Angiogenesis on coronary angiography is a marker for accelerated cardiac allograft vasculopathy as assessed by intravascular ultrasound. Clin Transplant. 2017;31:e13069.
31. Nikolova AP, Kobashigawa JA. Cardiac allograft vasculopathy: the enduring enemy of cardiac transplantation. Transplantation. 2019;103:1338–1348.
32. Loupy A, Toquet C, Rouvier P, et al. Late failing heart allografts: pathology of cardiac allograft vasculopathy and association with antibody-mediated rejection. Am J Transplant. 2016;16:111–120.
33. Duong Van Huyen JP, Fedrigo M, Fishbein GA, et al. The XVth Banff Conference on allograft pathology the banff workshop heart report: improving the diagnostic yield from endomyocardial biopsies and quilty effect revisited. Am J Transplant. 2020;20:3308–3318.
34. Hiemann NE, Knosalla C, Wellnhofer E, et al. Quilty in biopsy is associated with poor prognosis after heart transplantation. Transpl Immunol. 2008;19:209–214.
35. Hiemann NE, Knosalla C, Wellnhofer E, et al. Quilty indicates increased risk for microvasculopathy and poor survival after heart transplantation. J Heart Lung Transplant. 2008;27:289–296.
36. Mengel M, Sis B, Kim D, et al. The molecular phenotype of heart transplant biopsies: relationship to histopathological and clinical variables. Am J Transplant. 2010;10:2105–2115.
37. Huber W, Carey VJ, Gentleman R, et al. Orchestrating high-throughput genomic analysis with bioconductor. Nat Methods. 2015;12:115–121.
38. Famulski KS, de Freitas DG, Kreepala C, et al. Molecular phenotypes of acute kidney injury in human kidney transplants. J Am Soc Nephrol. 2012;23:948–958.
39. Halloran PF. T cell-mediated rejection of kidney transplants: a personal viewpoint. Am J Transplant. 2010;10:1126–1134.
40. Famulski KS, Broderick G, Einecke G, et al. Transcriptome analysis reveals heterogeneity in the injury response of kidney transplants. Am J Transplant. 2007;7:2483–2495.
41. Affymetrix. GeneChip® PrimeView™ Human Gene Expression Array data sheet. Available at http://www.affymetrix.com/support/technical/datasheets/primeview_array_cartridge_datasheet.pdf:1-4
. Accessed January 25, 2022.
42. Archer KJ, Mas VR, O’Brien TR, et al. Quality assessment of microarray data in a multicenter study. Diagn Mol Pathol. 2009;18:34–43.
43. Fan X, Lobenhofer EK, Chen M, et al. Consistency of predictive signature genes and classifiers generated using different microarray platforms. Pharmacogenomics J. 2010;10:247–257.
44. Klebanov L, Yakovlev A. How high is the level of technical noise in microarray data? Biol Direct. 2007;2:9.
45. Tambur AR, Campbell P, Claas FH, et al. Sensitization in transplantation: assessment of risk (STAR) 2017 working group meeting report. Am J Transplant. 2018;18:1604–1614.
46. Reed EF, Rao P, Zhang Z, et al. Comprehensive assessment and standardization of solid phase multiplex-bead arrays for the detection of antibodies to HLA-drilling down on key sources of variation. Am J Transplant. 2013;13:3050–3051.
47. Sis B, Jhangri GS, Bunnag S, et al. Endothelial gene expression in kidney transplants with alloantibody indicates antibody-mediated damage despite lack of C4d staining. Am J Transplant. 2009;9:2312–2323.
48. Einecke G, Sis B, Reeve J, et al. Antibody-mediated microcirculation injury is the major cause of late kidney transplant failure. Am J Transplant. 2009;9:2520–2531.
49. Halloran PF, Famulski KS, Chang J. A probabilistic approach to histologic diagnosis of antibody-mediated rejection in kidney transplant biopsies. Am J Transplant. 2017;17:129–139.
50. Parkes MD, Aliabadi AZ, Cadeiras M, et al. An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms. J Heart Lung Transplant. 2019;38:636–646.
51. Lombardi G, Sidhu S, Daly M, et al. Are primary alloresponses truly primary? Int Immunol. 1990;2:9–13.
52. Adams AB, Williams MA, Jones TR, et al. Heterologous immunity provides a potent barrier to transplantation tolerance. J Clin Invest. 2003;111:1887–1895.
53. Venner JM, Famulski KS, Badr D, et al. Molecular landscape of T cell-mediated rejection in human kidney transplants: prominence of CTLA4 and PD ligands. Am J Transplant. 2014;14:2565–2576.
54. Einecke G, Fairhead T, Hidalgo LG, et al. Tubulitis and epithelial cell alterations in mouse kidney transplant rejection are independent of CD103, perforin or granzymes A/B. Am J Transplant. 2006;6:2109–2120.
55. Nguyen LS, Ortuno S, Lebrun-Vignes B, et al. Transplant rejections associated with immune checkpoint inhibitors: a pharmacovigilance study and systematic literature review. Eur J Cancer. 2021;148:36–47.
56. Aubert O, Loupy A, Hidalgo L, et al. Antibody-mediated rejection due to preexisting versus de novo donor-specific antibodies in kidney allograft recipients. J Am Soc Nephrol. 2017;28:1912–1923.
57. Diebolder CA, Beurskens FJ, de Jong RN, et al. Complement is activated by IgG hexamers assembled at the cell surface. Science. 2014;343:1260–1263.
58. Böhmig GA, Eskandary F, Doberer K, et al. The therapeutic challenge of late antibody-mediated kidney allograft rejection. Transpl Int. 2019;32:775–788.
59. Koenig A, Mezaache S, Callemeyn J, et al. Missing self-induced activation of NK Cells combines with non-complement-fixing donor-specific antibodies to accelerate kidney transplant loss in chronic antibody-mediated rejection. J Am Soc Nephrol. 2021;32:479–494.
60. Koenig A, Chen CC, Marçais A, et al. Missing self triggers NK cell-mediated chronic vascular rejection of solid organ transplants. Nat Commun. 2019;10:5350.
61. Coutance G, Zouhry I, Racapé M, et al. Correlation between microvascular inflammation in endomyocardial biopsies and rejection transcripts, donor-specific antibodies, and graft dysfunction in antibody-mediated rejection. Transplantation. 2022;106:1455–1464.
62. Madill-Thomsen KS, Reeve J, et al. Assessing the relationship between molecular rejection and parenchymal injury in heart transplant biopsies. Transplantation. 2022;106:2205–2216.
63. Halloran PF, Madill-Thomsen K, Aliabadi-Zuckermann AZ, et al. Many heart transplant biopsies currently diagnosed as no rejection have mild molecular antibody-mediated rejection-related changes. J Heart Lung Transplant. 2022;41:334–344.
64. Halloran PF, Potena L, Van Huyen JD, et al. Building a tissue-based molecular diagnostic system in heart transplant rejection: the heart Molecular Microscope Diagnostic (MMDx) System. J Heart Lung Transplant. 2017;36:1192–1200.
65. Tourki B, Halade G. Leukocyte diversity in resolving and nonresolving mechanisms of cardiac remodeling. Faseb J. 2017;31:4226–4239.
66. Tham YK, Bernardo BC, Ooi JY, et al. Pathophysiology of cardiac hypertrophy and heart failure: signaling pathways and novel therapeutic targets. Arch Toxicol. 2015;89:1401–1438.
67. Montecucco F, Liberale L, Bonaventura A, et al. The role of inflammation in cardiovascular outcome. Curr Atheroscler Rep. 2017;19:11.
68. Frangogiannis NG. Inflammation in cardiac injury, repair and regeneration. Curr Opin Cardiol. 2015;30:240–245.
69. Glezeva N, Horgan S, Baugh JA. Monocyte and macrophage subsets along the continuum to heart failure: misguided heroes or targetable villains? J Mol Cell Cardiol. 2015;89:136–145.
70. Oka T, Hikoso S, Yamaguchi O, et al. Mitochondrial DNA that escapes from autophagy causes inflammation and heart failure. Nature. 2012;485:251–255.
71. Liu H, Liu X, Zhuang H, et al. Mitochondrial contact sites in inflammation-induced cardiovascular disease. Front Cell Dev Biol. 2020;8:692.
72. Bordon Y. Macrophages clean up to keep the heart pumping. Nat Rev Immunol. 2020;20:648–649.
73. McNally EM. Cardiac macrophages—keeping the engine running clean. N Engl J Med. 2020;383:2474–2476.
74. Hulsmans M, Sager HB, Roh JD, et al. Cardiac macrophages promote diastolic dysfunction. J Exp Med. 2018;215:423–440.
75. Nicolás-Ávila JA, Lechuga-Vieco AV, Esteban-Martínez L, et al. A network of macrophages supports mitochondrial homeostasis in the heart. Cell. 2020;183:94–109.e23.
76. Madill-Thomsen KS, Böhmig GA, Bromberg J, et al.; INTERCOMEX Investigators. Donor-specific antibody is associated with increased expression of rejection transcripts in renal transplant biopsies classified as no rejection. J Am Soc Nephrol. 2021;32:2743–2758.
77. Halloran PF, Madill-Thomsen KS, Pon S, et al.; INTERCOMEX Investigators. Molecular diagnosis of ABMR with or without donor-specific antibody in kidney transplant biopsies: differences in timing and intensity but similar mechanisms and outcomes. Am J Transplant. 2022;22:1976–1991.
78. Reeve J, Böhmig GA, Eskandary F, et al.; MMDx-Kidney study group. Assessing rejection-related disease in kidney transplant biopsies based on archetypal analysis of molecular phenotypes. JCI Insight. 2017;2:94197.
79. Halloran P, Reeve J, Grp IS. Molecular profiles of injury in heart transplant allograft: relationship to rejection, ejection fraction, and survival. Am J Transplant. 2020;20:315–316.
80. Einecke G, Reeve J, Mengel M, et al. Expression of B cell and immunoglobulin transcripts is a feature of inflammation in late allografts. Am J Transplant. 2008;8:1434–1443.
81. Einecke G, Reeve J, Gupta G, et al.; INTERCOMEX investigators. Factors associated with kidney graft survival in pure antibody-mediated rejection at the time of indication biopsy: importance of parenchymal injury but not disease activity. Am J Transplant. 2021;21:1391–1401.
82. Mengel M, Loupy A, Haas M, et al. Banff 2019 meeting report: molecular diagnostics in solid organ transplantation-consensus for the banff human organ transplant (B-HOT) gene panel and open source multicenter validation. Am J Transplant. 2020;20:2305–2317.
83. Adam N, Coutance G, Viailly PJ, et al. Reverse transcriptase multiplex ligation-dependent probe amplification in endomyocardial biopsies for the diagnosis of cardiac allograft rejection. J Heart Lung Transplant. 2020;39:115–124.
84. Agbor-Enoh S, Shah P, Tunc I, et al.; GRAfT Investigators. Cell-free DNA to detect heart allograft acute rejection. Circulation. 2021;143:1184–1197.
85. Tsuji N, Agbor-Enoh S. Cell-free DNA beyond a biomarker for rejection: biological trigger of tissue injury and potential therapeutics. J Heart Lung Transplant. 2021;40:405–413.
86. Shah P, Valantine HA, Agbor-Enoh S. Transcriptomics in transplantation: more than just biomarkers of allograft rejection. Am J Transplant. 2021;21:2000–2001.
87. Halloran PF, Reeve J, Madill-Thomsen KS, et al.; Trifecta Investigators. The Trifecta study: comparing plasma levels of donor-derived cell-free DNA with the molecular phenotype of kidney transplant biopsies. J Am Soc Nephrol. 2022;33:387–400.
88. Gupta G, Moinuddin I, Kamal L, et al. Correlation of donor-derived cell-free DNA with histology and molecular diagnoses of kidney transplant biopsies. Transplantation. 2022;106:1061–1070.
89. Reeve J, Böhmig GA, Eskandary F, et al.; INTERCOMEX MMDx-Kidney Study Group. Generating automated kidney transplant biopsy reports combining molecular measurements with ensembles of machine learning classifiers. Am J Transplant. 2019;19:2719–2731.