If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Reprint requests: Philip F. Halloran, MD, PhD, Alberta Transplant Applied Genomics Centre, #250 Heritage Medical Research Centre, University of Alberta, Edmonton, AB T6G 2S2, Canada. Telephone: 780-492-6160. Fax: 780-407-7450.
Improved understanding of lung transplant disease states is essential because failure rates are high, often due to chronic lung allograft dysfunction. However, histologic assessment of lung transplant transbronchial biopsies (TBBs) is difficult and often uninterpretable even with 10 pieces.
METHODS
We prospectively studied whether microarray assessment of single TBB pieces could identify disease states and reduce the amount of tissue required for diagnosis. By following strategies successful for heart transplants, we used expression of rejection-associated transcripts (annotated in kidney transplant biopsies) in unsupervised machine learning to identify disease states.
RESULTS
All 242 single-piece TBBs produced reliable transcript measurements. Paired TBB pieces available from 12 patients showed significant similarity but also showed some sampling variance. Alveolar content, as estimated by surfactant transcript expression, was a source of sampling variance. To offset sampling variation, for analysis, we selected 152 single-piece TBBs with high surfactant transcripts. Unsupervised archetypal analysis identified 4 idealized phenotypes (archetypes) and scored biopsies for their similarity to each: normal; T-cell‒mediated rejection (TCMR; T-cell transcripts); antibody-mediated rejection (ABMR)-like (endothelial transcripts); and injury (macrophage transcripts). Molecular TCMR correlated with histologic TCMR. The relationship of molecular scores to histologic ABMR could not be assessed because of the paucity of ABMR in this population.
CONCLUSIONS
Molecular assessment of single-piece TBBs can be used to classify lung transplant biopsies and correlated with rejection histology. Two or 3 pieces for each TBB will probably be needed to offset sampling variance.
The Registry of the International Society for Heart and Lung Transplantation: Thirty-second official adult lung and heart‒lung transplantation report—2015; Focus theme: Early graft failure.
with significant graft loss due to the poorly understood syndrome of chronic lung allograft dysfunction (CLAD). The extent to which T-cell‒mediated rejection (TCMR, also known as acute cellular rejection) and antibody-mediated rejection (ABMR) contribute to CLAD is not known. The standard of care (SOC) for diagnosing lung rejection is the transbronchial biopsy (TBB) interpreted using International Society for Heart and Lung Transplantation (ISHLT) guidelines.
(B-grade), both likely reflecting TCMR, and obliterative bronchiolitis (C-grade). The LARGO study of TBBs showed that sub-optimal sampling was common and interobserver agreement was poor.
In kidney and heart transplants, a microarray-based diagnostic system (Molecular Microscope® Diagnostic System, or MMDx) has been developed to detect TCMR and ABMR
with high technical reproducibility while requiring less tissue. For heart transplants, we used unsupervised machine learning to discover phenotypes based on rejection-associated transcripts (RATs) shared between kidney and heart transplants.
: S1normal (absence of inflammation/rejection); S2TCMR (effector T-cell transcripts); S3ABMR (mainly endothelial transcripts); and S4injury (macrophage transcripts).
The present prospective observational study (INTERLUNG ClinicalTrials.gov: NCT02812290) adapted the strategy developed for heart biopsies to single-piece lung transplant TBBs (a restriction imposed by institutional review boards [IRBs]). The research plan is outlined in Figure 1. Our goals were to use a discovery-based approach to determine whether TBB pieces can be analyzed by microarrays, to use unsupervised analysis to identify rejection and injury in TBBs, and to relate these findings to histologic diagnoses and donor-specific antibody (DSA) status.
We prospectively enrolled consenting adult lung transplant recipients from 7 centers (refer to Table S1 in the Supplementary Material available online at www.jhltonline.org/). We collected indication biopsies (i.e., clinician concern), protocol biopsies, and follow-up biopsies, collected as per local SOC (see Methods for clinical diagnostics in the Supplementary Material online).
Lung biopsies
TBBs were performed in an accessible lobe as per SOC at participating institutions, with up to 10 TBB pieces collected per patient, depending on safety and tolerance. One TBB piece was stabilized in RNAlater for microarray analysis (see Methods in Supplementary Material online), and all other pieces were sent for SOC histologic assessment. Mucosal biopsies from the second bronchial bifurcation (2BMB) were performed in a subset of cases to define differentially expressed genes in TBBs. Biopsies were processed with microarrays and the data were normalized, as described elsewhere.
(ABMR-RATs). The union of the top 200 probe sets from each comparison yielded 453 RAT probe sets (260 unique genes), which were used for unsupervised analyses.
Data analyses were performed using Bioconductor version 3.6 and R version 3.4.2. Top genes differentially expressed in TBBs vs 2BMBs were identified by comparing mean probe set expression using Bayesian t-tests with the R package “limma.” Principal component analysis (PCA) was performed based on RAT expression using the R package “FactoMineR.” Archetypal analysis (AA) was performed using the “archetypes” package.
AA is a method of unsupervised analysis that extrapolates a pre-defined k number of theoretical biopsies (archetypes, denoted by “A”) that represent the idealized phenotypes in a given data set. Each biopsy is then is scored (denoted by “S”) to reflect its weighted distance from each k theoretical archetype such that the sum of scores is 1:
We performed AA on TBBs based on RAT expression and assigned TBBs to groups according to highest archetype score.
This study was approved the University of Alberta Research Ethics Board (Pro00048176) and the IRBs at participating centers. Written informed consent was received from patients before inclusion in this study.
Results
Patient and biopsy characteristics
We obtained 242 single-piece TBBs from 209 lung transplant recipients at 7 centers, with a second piece from 12 patients for studying sampling variation (Table 1, see also Table S2 online). TBB histology was interpreted as per SOC at the biopsying center. TBBs were done for clinical indications (e.g., decline in spirometry), follow-up, and protocol/surveillance; 69 biopsies (30% of tested patients) were from patients positive for DSA (Table 1). Most TBBs were done early post-transplant (median 256 days).
Table 1INTERLUNG Transbronchial Biopsy Histologic Characteristics and Patient DSA Status
The most recent DSA status at time of recent biopsy was used, if known. DSA statuses dated more than 14 days after the biopsy were not considered. If the most recent DSA status at time of biopsy was not known, but the patient was most recently PRA negative, the DSA status was presumed negative. ACR, acute cellular rejection; AMR, antibody-mediated rejection; DSA, donor-specific antibody; ISHLT, International Society for Heart and Lung Transplantation; PRA, panel-reactive antibody; TBB, transbronchial biopsy; TCMR, T-cell‒mediated rejection.
Not performed in the center's standard-of-care TBB assessment.
87 (36%)
74 (49%)
Not recorded
0
0
a The most recent DSA status at time of recent biopsy was used, if known. DSA statuses dated more than 14 days after the biopsy were not considered. If the most recent DSA status at time of biopsy was not known, but the patient was most recently PRA negative, the DSA status was presumed negative. ACR, acute cellular rejection; AMR, antibody-mediated rejection; DSA, donor-specific antibody; ISHLT, International Society for Heart and Lung Transplantation; PRA, panel-reactive antibody; TBB, transbronchial biopsy; TCMR, T-cell‒mediated rejection.
b Diagnoses described by the local pathologist. Some samples had multiple diagnoses.
c Samples were of insufficient quality to make a diagnosis based on histologic assessment (ungradable).
d Not performed in the center's standard-of-care TBB assessment.
In 234 TBBs with available ISHLT grades, 144 were A0, 56 were A-grade ≥1 (i.e., ACR or TCMR), and 34 were ungradable (Ax), similar to previous studies.
ISHLT B- and C-grade lesions could not be read in half of the TBBs and, of those that were gradable, only 9 were graded B ≥1R and 6 C ≥1, similar to previous studies,
and 107 biopsies (46%) came from centers where C-grade assessment is not SOC. Thus, there were too few TBBs with B and C lesions to analyze these lesions’ relationships with molecular states.
Overall pathology diagnoses were: ACR, 50 cases; antibody-mediated rejection (AMR), 12 cases; and mixed rejection, 5 cases. There were 12 clinically diagnosed AMR cases, but only 5 of those were called AMR by histology, consistent with the known diagnostic uncertainty with respect to AMR in lung transplants.
Agreement between TBB pieces taken at the same time
In general, single-piece TBBs yielded high-quality RNA suitable for microarray analysis. In 12 cases where 2 TBB pieces were available from the same biopsy, we assessed agreement between the 2 pieces in terms of expression of PBTs representing biologic processes in rejection and injury (see Methods). In a permutation test of 1,000 iterations of random pairing, the true TBB pairs showed better agreement in PBT scores than expected by chance (p < 0.01; see Figure S1 online), but with some variation in individual PBT scores between TBBs from the same patient.
TBB heterogeneity in alveolar content
Because alveolar content varies between single TBB pieces, we estimated alveolar content by identifying transcripts characteristic of alveoli. We compared gene expression in 242 TBBs with 34 biopsies from the mucosa at the second bronchial bifurcation (2BMB), which lack alveoli (Table 2). Most of the top differentially expressed transcripts were surfactant transcripts (SFTs), expressed up to 700 times more strongly in TBBs than 2BMBs.
Table 2Top Differentially Expressed Genes in TBB vs 2BMB
We selected highly expressed SFTs to reflect alveolar content. Twenty-three TBBs (10%) had no SFTs (mean expression <100) and presumably lacked alveoli, and some others had low SFTs (Figure 2). Alveolar content could affect endothelial transcript expression because alveoli contain more capillaries than non-alveolar tissues.
Figure 2Surfactant (SFT) transcript expression in 242 TBBs. The 50 highest variance probe sets in 242 TBBs were identified. From this, 11 probe sets representing 4 SFT transcripts were identified (11757270_x_at, 11763961_x_at, 11754641_x_at, 11742494_s_at, 11735664_s_at, 11764024_x_at, 11748373_s_at, 11734773_x_at, 11745166_x_at, 11763809_x_at, 11749911_x_at) and their geometric mean expression across 242 TBB samples was calculated. The samples were ordered by decreasing geometric mean. The 152 samples to the left of the dashed vertical line were considered to have sufficiently high SFT expression (i.e., high alveolar content) to be used in subsequent analyses of the TBBs.
To reduce heterogeneity in alveolar content, we restricted the subsequent analyses to 152 TBBs with SFT expression level >12,380, the point where mean SFT expression dropped sharply (Figure 2). Lung function at biopsy (as measured by forced expiratory volume in 1 second [FEV1]) did not differ significantly between the excluded biopsies and the 152 high-SFT biopsies (p = 0.49).
AA and PCA in 152 high-surfactant TBBs
Following the strategy successful for heart transplant biopsies,
we used unsupervised AA based on RAT expression to derive the TBB archetype model. RAT performance was validated in the TBBs: ABMR RATs and TCMR RATs correlated opposingly with PC2, and Rejection RATs correlated with PC1 (Figure 3). The scree plot in Figure 4A illustrates how much variance is explained by adding additional archetypes to the model. Based on this and our experience with heart biopsies
we elected to use 4 archetypes, provisionally designated as A1 (normal), A2 (TCMR), A3 (ABMR-like, reflecting the uncertainty in conventional lung AMR diagnosis), and A4 (injury).
As in hearts, we use “injury” to encompass all processes causing abnormal RAT expression uncorrelated with the rejection scores S2TCMR and S3ABMR-like. Each TBB was assigned 4 corresponding scores (S1normal, S2TCMR, S3ABMR-like, and S4injury), reflecting its weighted similarity to each archetype.
Figure 3Correlations between RATs and the principal component scores in principal component analysis of 152 high-surfactant TBBs. The 453 RATs are the union of the top 200 transcripts associated with each of the following class comparisons in kidney transplant biopsies: all rejection (rej) vs everything else; TCMR vs everything else; and ABMR vs everything else. The transcripts are color-coded according to which of the 3 comparisons they were associated with.
Figure 4Archetypal analysis of 152 TBBs based on expression of rejection-associated transcripts (RATs). (A) Scree plot depicting residual sums of squares for archetypal analysis models built using different numbers of archetypes, k. In models with k = 5 through k = 10 archetypes, the relatively stable residual sum of squares suggests that little information is gained by examining more than 4 archetypes (red dot). Thus, the present study focuses on a 4-archetype model. (B‒D) RAT-based archetypal analysis of 152 TBBs using 4 archetypes, visualized in RAT-based principal component analysis (PCA). (B) plots principal component 2 (PC2) vs PC1, (C) plots PC2 vs PC3, and (D) plots PC3 vs PC1. Each sample is colored according to its archetype cluster membership, which was determined by its highest of 4 archetype scores. The large ghosted points labeled “A1,” “A2,” “A3,” and “A4” denote the position of the theoretical archetypes in PCA.
TBBs were plotted in RAT-based PCA (Figure 4B‒D), giving each biopsy linearly uncorrelated principal component scores (PC1, PC2, and PC3) that describe the principal sources of variation among the biopsies. TBBs were grouped and colored according to their highest AA score (Figure 4B‒D). PC1 separated rejection and injury from normal (Figure 4B and D); PC2 separated ABMR-like TBBs from TCMR (Figure 4B and C); and PC3 separated injury from the rejection and normal groups (Figure 4C and D).
Correlation of AA and PCA scores with PBT expression
S1normal and PC1 demonstrated strong correlations with expression of all-rejection RATs, TCMR-RATs, and interferon-gamma (IFNG)-inducible transcripts (GRIT1s) (Figure 5). S1normal correlations were strongly negative, whereas PC1 correlations were strongly positive. S2TCMR correlated with cytotoxic T-cell transcripts (QCATs) and TCMR-RATs. S3ABMR-like and PC2 correlated with ABMR-associated transcript sets DSASTs, ENDATs, and ABMR-RATs. S4injury and PC3 correlated most strongly with macrophage transcripts (QCMATs), are associated with injury and depressed left ventricular fraction in heart transplants.
Figure 5Correlations of molecular scores in 152 TBBs. Spearman correlations between molecular features of the samples are given and the cells are colored according to the magnitude of the correlation. Archetype scores (S1normal, S2TCMR, S3ABMR-like, S4injury) and principal component scores (PC1, PC2, PC3) were trained on RAT expression in 152 TBBs. QCAT, ENDAT, DSAST, QCMAT, and GRIT1 transcripts in the top 50% of expression variance in the full set of 242 TBBs were used. RATs were excluded from all other transcript sets. Column order is dictated by Ward's minimum variance clustering on a Manhattan distance matrix of the scores across 152 TBBs. ABMR-RAT, ABMR-associated RATs; TCMR-RAT, TCMR-associated RATs; Rejection-RAT, rejection-associated RATs; SFT, surfactant; DSAST, DSA-associated transcripts; ENDAT, endothelium-associated transcripts; QCMAT, macrophage-associated transcripts; QCAT, cytotoxic T-cell‒associated transcripts; GRIT1, interferon-gamma‒inducible transcripts.
SFT expression did not correlate with endothelial and ABMR-related PBTs in TBBs with adequate alveoli, validating our strategy of limiting analyses to high-surfactant TBBs.
Top transcripts correlated with AA and PCA scores
Using RAT expression as the basis for PCA and archetype scores does not guarantee that RATs will be the dominant transcripts associated with these scores. The top 20 unique transcripts by absolute correlation with each score (out of 50,000 probe sets on the microarray) are summarized in terms of their expression in a human cell panel
in Table 3, and listed in Tables S3 through S6 (archetype scores) and Tables S7 through S9 (PCA scores) online. Thirteen of 20 S1normal, 18 of 20 S2TCMR, 2 of 20 S3ABMR-like, 0 of 20 S4injury, 15 of 20 PC1, 3 of 20 PC2, and 1 of 20 PC3 top associated transcripts were RATs. S1normal and PC1 anti-correlated and correlated, respectively, with inflammation; S2TCMR correlated with effector T-cell transcripts and certain macrophage transcripts (e.g., see ADAMDEC1 in Table S4 online); S3ABMR-like correlated with endothelial transcripts; and S4 correlated with macrophage transcripts. A full analysis of these relationships will be presented in a future study (manuscript in preparation).
Table 3Summary of Transcripts Associated With Archetype Scores and Principal Components
The only histologic diagnosis represented sufficiently in this TBB population for comparison with MMDx scores was A-grade histologic TCMR/ACR. Elevated S2TCMR scores were strongly associated with A ≥1 (one-tailed Mann‒Whitney U-test, p = 1 × 10−5). High PC1 (p = 1 × 10−3) and low PC2 (p = 8 × 10−4) scores were also associated with A ≥1. S3ABMR-like, S4injury, and PC3 were not significantly associated with A-grade (p > 0.05). There were not enough samples with positive B- or C-grades to assess relationships.
As expected, given the low frequency of ABMR in this population, there were no associations between molecular scores and either clinical or histologic diagnoses of ABMR or with DSA at biopsy. A class comparison of biopsies from DSA-positive vs DSA-negative patients revealed no transcripts significantly associated with DSA positivity (data not shown).
Relationships between molecular scores and infection
Many biopsies came from patients recorded by the clinician as having some evidence for infection, but these were highly heterogeneous and did not differ significantly between archetype groups (Chi-square test: any infection, p = 0.33; viral, p = 0.60; bacterial, p = 0.40; fungal, p = 0.50).
Discussion
This prospective discovery study of consented patients used unsupervised analysis based on RAT expression to identify molecular states in lung transplant TBBs. The goal was not a comprehensive diagnostic system, which would require a larger biopsy set, but rather an exploration of the feasibility and utility of using molecular approaches previously used for heart transplants, leveraging the sharing of molecular features of rejection and injury between organs in related disease processes, as previously demonstrated for heart
IRBs limited the study mainly to single-piece TBBs, forcing us to confront TBB composition heterogeneity. Every single-piece TBB produced high-quality readable RNA. Second TBB pieces available for 12 TBBs showed significant similarity in PBT expression, but with considerable variability. One source of variability was alveolar content, requiring that the analyses be restricted to 152 single-piece TBBs with high SFT expression. We performed unsupervised analyses of these TBBs to assign scores analogous to those used for heart transplants, and demonstrated the relationship between these scores and histologic TCMR as reflected by the A lesions, setting a framework for future clinical development of a definitive diagnostic system.
The S2TCMR score emerges as a reliable indicator of true TCMR, supported by its molecular similarity to TCMR in kidney and heart transplant biopsies and its association with histologic A-grade TCMR. For example, ADAMDEC1, which encodes an enzyme expressed by activated macrophages, is highly associated with S2TCMR in lungs as well as with TCMR in kidney
and heart transplants (unpublished data). ADAMDEC1 is not prominent in TBBs with high S4injury and PC3 scores, despite expression of many other macrophage transcripts. We believe that ADAMDEC1 induction reflects a macrophage activation phenotype relatively specific for the cognate TCMR process in kidney, heart, and lung, reflecting interactions between effector T cells and macrophages, distinct from non-specific inflammation.
The relationship of S3ABMR-like scores to true ABMR, like many aspects of ABMR in lung transplants, remains elusive and will be resolved in future studies targeting ABMR. Given the paucity of ABMR in the current set of TBBs, our finding that neither S3ABMR-like scores nor individual transcripts correlate with DSA positivity was expected. The association of DSA with conventional lung transplant phenotypes has also been difficult to establish.
The non-rejection molecular changes that we designated injury—meaning all non-rejection sources of disturbance in RAT expression—are of particular interest in lung transplantation because of the many non‒rejection-related stresses on lung tissue.
We first explored the molecular injury phenotype in kidney transplants, mapping transcripts (including some macrophage transcripts) associated with impaired function (glomerular filtration rate) in the absence of rejection.
In TBBs, like heart transplants, S4injury and PC3 were strongly associated with macrophage transcripts. Although we found no correlation between S4injury scores and FEV1 in lungs, this issue will be of interest in future studies. Association of injury with macrophages does not exclude a role for neutrophils in lung injury: mRNA studies underestimate the role of neutrophils because they are unstable in vitro and cannot be included in cell panels. Moreover, granulocyte populations have abundant RNAses that rapidly destroy their mRNA. Despite this, the top transcript associated with S4injury was neutrophil cytosolic factor 2 (NCF2), a transcript shared by neutrophils and macrophages. Pathway analysis also pointed to neutrophil involvement (data not shown). The role of neutrophils in lung injury is of particular interest because lung tissue is characterized by high neutrophil influx compared with hearts or kidneys.
The ability to molecularly phenotype small pieces from TBB opens the possibility of establishing the mechanisms of CLAD and being able to predict its development. CLAD is difficult to study because of its variable presentation and lack of histologic definition, and also because it may involve lung compartments that are underrepresented in TBBs (e.g., small airways). The MMDx TBB approach can shed light on the role of rejection or related inflammatory mechanisms in CLAD.
A major advantage of the MMDx system is the assignment of continuous measurements with high technical reproducibility.
The significant similarity between single TBBs is encouraging evidence that they represent the biology of the lung, but the variability between single TBB pieces suggests that reliable MMDx diagnoses will require at least 2 pieces on 1 microarray. As algorithms mature with more samples, sampling variation (including alveolar content) may be reduced by incorporating a correction factor because we have been able to do this for the cortex‒medulla issue in kidneys.
Our study has provided a definitive basis for the next steps in the discovery process: a new INTERLUNG study focusing on building the reference set with a particular focus on the molecular definition of ABMR and CLAD, which will require long-term follow-up. Our results support our request to IRBs to permit 2 TBB pieces per biopsy to offset sampling heterogeneity. Ultimately, our goal is to develop a reference set of at least 1,000 TBBs and to relate the findings to clinical phenotypes, effects of treatment, and prognosis. This will allow us to derive a set of RATs in the TBBs, which will in turn permit us to validate the use of kidney RATs in the analysis of lung rejection; compare the lung, heart, and kidney transcripts; and potentially identify molecular processes that may be unique to rejection in lung tissue. In parallel, we are now in the process of exploring the utility of bronchial mucosal biopsies, a safer biopsy format (ClinicalTrials.gov NCT02812290).
Molecular detection of rejection–like changes in proximal bronchial mucosal lung transplant biopsies: initial findings of the INTERLUNG study [abstract].
P.F.H. holds shares in Transcriptome Sciences, Inc., a University of Alberta research company with an interest in molecular diagnostics. The remaining authors have no conflicts of interest to disclose.
This research was supported by funds and/or resources from the Mendez National Institute of Transplantation Foundation, Roche Organ Transplant Research Foundation, and the University of Alberta Hospital Foundation.
The authors are grateful to Drs. Justin Weinkauf, Dale Lien, Ali Kapasi, and Alim Hirji for providing biopsy specimens.
Portions of this study were presented at the 38th annual meeting and scientific sessions of the International Society for Heart and Lung Transplantation, April 12, 2018, Nice, France.
The Registry of the International Society for Heart and Lung Transplantation: Thirty-second official adult lung and heart‒lung transplantation report—2015; Focus theme: Early graft failure.
Molecular detection of rejection–like changes in proximal bronchial mucosal lung transplant biopsies: initial findings of the INTERLUNG study [abstract].
Long-term survival after lung transplantation remains limited by the development of chronic lung allograft dysfunction (CLAD), which affects approximately 50% of lung recipients by 5 years post-transplant.1 Given the lack of a proven therapy once CLAD is diagnosed, risk factor mitigation is a key preventive strategy. The principal risk factor for CLAD is A-grade acute cellular rejection (ACR). A-grade ACR is diagnosed by transbronchial biopsy (TBB), exhibiting perivascular mononuclear cell infiltrates that extend into the interstitium with higher grade rejection.
31347-6&id=gr1.jpg){kind=link}
31347-6&id=gr2.jpg){kind=link}
31347-6&id=gr3.jpg){kind=link}
31347-6&id=gr4.jpg){kind=link}
31347-6&id=gr5.jpg){kind=link}