Molecular Landscape of T Cell–Mediated Rejection in Human Kidney Transplants: Prominence of CTLA4 and PD Ligands
Abstract
We used expression microarrays to characterize the changes most specific for pure T cell–mediated rejection (TCMR) compared to other diseases including antibody-mediated rejection in 703 human kidney transplant biopsies, using a Discovery Set–Validation Set approach. The expression of thousands of transcripts—fold change and association strength—changed in a pattern that was highly conserved between the Discovery and Validation sets, reflecting a hierarchy of T cell signaling, costimulation, antigen-presenting cell (APC) activation and interferon-gamma (IFNG) expression and effects, with weaker associations for inflammasome activation, innate immunity, cytotoxic molecules and parenchymal injury. In cell lines, the transcripts most specific for TCMR were expressed most strongly in effector T cells (e.g. CTLA4, CD28, IFNG), macrophages (e.g. PDL1, CD86, SLAMF8, ADAMDEC1), B cells (e.g. CD72, BTLA) and IFNG-treated macrophages (e.g. ANKRD22, AIM2). In pathway analysis, the top pathways included T cell receptor signaling and CTLA4 costimulation. These results suggest a model in which TCMR creates an inflammatory compartment with a rigorous hierarchy dominated by the proximal aspects of cognate engagement of effector T cell receptor and costimulator triggering by APCs. The prominence of inhibitors like CTLA4 and PDL1 raises the possibility of active negative controls within the rejecting tissue.
Abbreviations
-
- ABMR
-
- antibody-mediated rejection
-
- AKI
-
- acute kidney injury
-
- APC
-
- antigen-presenting cell
-
- DC
-
- dendritic cell
-
- FDR
-
- false discovery rate
-
- HUVECs
-
- human umbilical vein endothelial cells
-
- IFNG
-
- interferon-gamma
-
- IQR
-
- interquartile range
-
- NK
-
- natural killer
-
- PBMCs
-
- peripheral blood mononuclear cells
-
- RPTECs
-
- renal proximal tubular epithelial cells
-
- TCMR
-
- T cell–mediated rejection
Introduction
The molecular biology of T cell–mediated rejection (TCMR) of organ transplants is of interest not only for transplant immunology but also as a model for T cell effector mechanisms operating in autoimmunity and in cancer immunotherapy with agents directed against CTLA4 and PD1 1. TCMR is characterized by interstitial infiltration by effector T cells, myeloid cells with macrophage and dendritic cell (DC) markers, and B cells. Based on expression of cytotoxic molecules such as perforin (PRF1), granzymes A and B (GZMA, GZMB), granulysin and Fas ligand (CD95L) 2, 3, the mechanism of tissue injury in TCMR was at one time considered to be direct killing of individual donor cells. However, studies in KO mice demonstrated that TCMR was independent of the principal cytotoxic molecules 4. The current concept is that TCMR is an inflammatory reaction within the interstitium of the allograft tissues triggered by the cognate engagement of donor antigen on antigen-presenting cells (APCs) by a small number of primed effector T cells and heterologous memory T cells 5, 6. The resulting synapse activates the effector T cell, inducing interferon-gamma (IFNG) production 7, and activates the APC. The signals from this event recruit myeloid cells and activate the inflammasome 8, creating an interstitial inflammatory compartment. The parenchyma dedifferentiates, manifesting gene expression changes similar to those in acute kidney injury (AKI) 9, while losing molecules associated with specialized organ functions such as solute carriers 10.
The study of human TCMR is complicated by the occurrence of antibody-mediated rejection (ABMR) in the same population. ABMR is an intracapillary process distinct from the tubulo-interstitial process of TCMR, they use different cognate antigen recognition processes, yet share many molecular features 11, 12. Thus comparison of TCMR to ABMR is useful for defining the most TCMR-specific changes. The unique features of TCMR versus ABMR must be studied in human tissues because mouse and rat kidney transplants do not develop typical ABMR 13. Until recently, human studies were limited by the inability of conventional methods to separate pure TCMR from ABMR reliably, because the majority of ABMR was C4d-negative and missed by the prevailing histology criteria 14. Consequently studies of “rejection” probably included undetected ABMR 15-17. However, we developed a new histology classification that separates TCMR from ABMR, the ATAGC Reference Standard classification (http://atagc.med.ualberta.ca/, and published in references 18, 19), anticipating the recently proposed Banff revised criteria for rejection 20. This allowed us to develop a microarray test for TCMR using a molecular classifier, developed in a Discovery Set of 403 biopsies 18 and confirmed in a Validation Set of 300 biopsies 19.
In the present study we took advantage of the presence of ABMR and other diseases in the transplant population to highlight the changes most specific for TCMR versus other disease and injury processes. We hypothesized that the transcript changes most strongly associated with pure TCMR would reflect cognate effector T cell–APC engagement in the interstitial inflammatory compartment. We started with the 30 TCMR classifier transcripts identified in the Discovery Set, examined the robustness of these relationships in the Validation Set, then mapped downstream consequences such as IFNG effects, previously described rejection biomarkers, and transcripts associated with inflammasome activation and parenchymal injury. We used cell lines and pathway analysis to explore the underlying biology of the transcripts most specific for pure TCMR.
Materials and Methods
Patient population and biopsy collection
Two cohorts of kidney transplant biopsies were studied: the Discovery Set (403 biopsies from 315 patients) 18 and the Validation Set (300 biopsies from 264 patients) from six kidney transplant centers 19 (Table 1). Biopsies were characterized using the histology-based ATAGC Reference Standard that incorporated C4d-negative ABMR (left panel http://atagc.med.ualberta.ca/ and references 18, 19).
| Patient characteristics | Combined Set (patients = 579) | Discovery Set (patients = 315) | Validation Set (patients = 264) |
|---|---|---|---|
| Mean recipient age (years) | 48 (10–86) | 48 (14–81) | 49 (10–86) |
| Recipient gender (% male) [n = 579] | 375 (65%) | 197 (63%) | 178 (67%) |
| Race [n = 579] | |||
| Caucasian | 431 (74%) | 214 (68%) | 217 (82%) |
| Black | 45 (8%) | 33 (10%) | 12 (5%) |
| Other | 75 (13%) | 47 (15%) | 28 (11%) |
| NA | 28 (5%) | 21 (7%) | 7 (3%) |
| Primary disease [n = 579] | |||
| Diabetic nephropathy | 95 (16%) | 66 (21%) | 29 (11%) |
| Hypertension/large vessel disease | 38 (7%) | 29 (9%) | 9 (3%) |
| Glomerulonephritis/vasculitis | 154 (27%) | 119 (38%) | 35 (13%) |
| Interstitial nephritis/pyelonephritis | 27 (5%) | 19 (6%) | 8 (3%) |
| Polycystic kidney disease | 64 (11%) | 45 (14%) | 19 (7%) |
| Others | 40 (7%) | 15 (5%) | 25 (9%) |
| Unknown etiology | 161 (28%) | 22 (7%) | 139 (53%) |
| Mean donor age (years) | 41 (1–75) | 40 (2–69) | 41 (1–75) |
| Donor gender (% male) | 223 (39%) | 121 (38%) | 102 (39%) |
| Donor type (% deceased donor transplants) | 314 (54%) | 152 (48%) | 162 (61%) |
| Clinical characteristics at time of biopsy | Combined Set (biopsies = 703) | Discovery Set (biopsies = 403) | Validation Set (biopsies = 300) |
|---|---|---|---|
| Median and range time from transplant to biopsy (months) | 19 (0–427) | 17 (0.2–428) | 25 (0.1–330) |
| Indication for biopsy | |||
| Primary nonfunction | 11 (2%) | 10 (2%) | 1 (0%) |
| Rapid deterioration of graft function | 161 (23%) | 96 (24%) | 65 (22%) |
| Slow deterioration of graft function | 216 (31%) | 150 (37%) | 66 (22%) |
| Stable impaired graft function | 88 (13%) | 71 (18%) | 17 (6%) |
| Investigate proteinuria | 103 (15%) | 38 (9%) | 65 (22%) |
| Follow-up from previous biopsy | 20 (3%) | 14 (3%) | 6 (2%) |
| Others | 76 (11%) | 9 (2%) | 67 (22%) |
| Indication unknown | 28 (4%) | 15 (4%) | 13 (4%) |
| Maintenance immunosuppressive regimens at biopsy | |||
| MMF, tacrolimus, steroid | 284 (40%) | 176 (44%) | 108 (36%) |
| MMF, cyclosporine, steroid | 126 (18%) | 101 (25%) | 25 (8%) |
| Others | 293 (42%) | 126 (31%) | 167 (56%) |
| Conventional biopsy diagnosis | Combined Set (biopsies = 703) | Discovery Set (biopsies = 403) | Validation Set (biopsies = 300) |
|---|---|---|---|
| Acute kidney injury (AKI) | 64 (9%) | 50 (12%) | 14 (5%) |
| Polyomavirus nephropathy (PVN) | 25 (4%) | 12 (3%) | 13 (4%) |
| TCMR | 67 (10%) | 35 (9%) | 32 (11%) |
| Borderline | 89 (13%) | 43 (11%) | 46 (15%) |
| C4d-negative ABMR | 80 (11%) | 53 (13%) | 27 (9%) |
| C4d-positive ABMR | 30 (4%) | 17 (4%) | 13 (4%) |
| Mixed | 28 (4%) | 22 (5%) | 6 (2%) |
| Transplant glomerulopathy (TG) | 27 (4%) | 7 (2%) | 20 (7%) |
| Glomerulonephritis (GN) | 81 (12%) | 41 (10%) | 40 (13%) |
| Interstitial fibrosis and tubular atrophy (IFTA) | 72 (10%) | 34 (8%) | 38 (13%) |
| No major abnormalities (NOMOA) | 116 (17%) | 73 (18%) | 43 (14%) |
| Other | 24 (3%) | 16 (4%) | 8 (3%) |
- ABMR, antibody-mediated rejection; MMF, mycophenolate mofitil; TCMR, T cell-mediated rejection
All biopsies were obtained for clinical indications, usually dysfunction or proteinuria, as standard of care from 3 days to 35 years posttransplant. The study was approved by the institutional review board in each participating center.
Biopsies were processed for microarray analysis as described previously 18. One 18-gauge biopsy core was placed immediately into RNALater (Affymetrix, Santa Clara, CA). RNA extraction, labeling and hybridization to HG-U133 Plus 2.0 GeneChip arrays (Affymetrix) were carried out according to manufacturer's protocols (www.affymetrix.com). Microarrays were scanned, .CEL files were obtained using Gene Chip Operating Software 1.4.0 (Affymetrix), and robust multiarray averaging was used to normalize the microarrays 18. (The term transcript is used more or less synonymously for probe sets.)
Analyses and graphics were done using the “R” software package, version 2.15.2 (64-bit) 21 with various libraries from Bioconductor 2.8 22, or Microsoft Excel version 14 (Redmond, WA). Significance of probe set expression is expressed as an unadjusted p-value (Bayesian t-test), unless specified, where the false discovery rate (FDR) was used. Microarray expression files are posted on the Gene Expression Omnibus website for the Discovery (GSE36059) and Validation (GSE48581) Sets, and will be posted for the cell lines upon publication.
Selection of TCMR classifier transcripts
As previously published 18, a Discovery Set classifier algorithm compared TCMR biopsies to all others using t-tests in a random set of 90% of the 403 biopsies, selecting the top 20 transcripts by p-value. This algorithm was run 1000 times in random subsets of 90% of biopsies, and the 30 most frequently selected transcripts were designated the TCMR classifier transcripts in the Discovery Set 18. Biopsies with ambiguous diagnostic labels, borderline and mixed, were withheld from the training set 18. They were also withheld from the study presented here for the purpose of the study of pure TCMR.
Expression of transcripts in cultured cells
We isolated peripheral blood mononuclear cells (PBMCs) from whole blood of healthy volunteers by density gradient centrifugations using Ficoll® (GE Healthcare Life Sciences, Baie-D'Urfé, Quebec, Canada) and then purified the following cell populations for expression analysis on HG-U133 Plus 2.0 GeneChip arrays. (RNA for the cultured cells was prepared in the same manner as the biopsies.)
Effector T cells
CD4+ and CD8+ T cells from three and five healthy donors, respectively, were generated through repeated cycles of allostimulation starting with PBMCs cultured at a ratio of 3:1 with mitomycin C (Sigma, St. Louis, MO)-treated chronic myelogenous leukemia B cells (RPMI8866, ATCC). Recombinant human IL-2 (eBioscience, San Diego, CA) was added to the cultures at 50 U/mL and cultured for 5 days per round. After four rounds of stimulation, live cells were collected by Ficoll® density gradient centrifugation, followed by CD4+ and CD8+ cell purification using Easy Sep® negative selection kits (Stem Cell Technologies, Vancouver, BC, Canada) according to manufacturer instructions. Cell purity varied between 92% and 98% (assessed by flow cytometry). Effector phenotype was demonstrated by intracellular staining: 95 ± 3% of CD8+ T cells stained positive for GZMB after the final stimulation, and 96 ± 2% of CD4+ and 90 ± 3% of CD8+ T cells stained positive for IFNG upon restimulation 23.
B cells and natural killer cells
B and natural killer (NK) cells were purified from PBMCs using EasySep® negative selection kits (Stem Cell Technologies). Both cell populations remained unstimulated until the time of RNA extraction. B cells were >97% CD19+ and NK cells varied between 90% and 98% CD56+CD3−. Human NK cells were selected from donors with similar high ratios of CD56lo/CD56bright NK cells, suggestive of a cytolytic phenotype 24. The majority (average 96.1%) of NK cells showed a cytotoxic phenotype (CD56dim) as expected in whole blood 25.
Monocytes
Monocytes were isolated using EasySep® Human CD14+ Selection Kit (Stem Cell Technologies) direct from the PBMCs.
Macrophages
Monocytes were resuspended in complete RPMI, allowed to adhere on 100 mm plates (BD Falcon, Missassauga, Ontario, Canada), and left for 24 h, with or without recombinant human IFNG (500 U/mL; eBioscience).
DCs
Monocytes were cultured for 7 days in the presence of IL-4 (500 U/mL; eBioscience) and recombinant human granulocyte-macrophage colony stimulating factor (500 U/mL; eBioscience). DCs were either immature (“Immature DCs”), or they were matured with lipopolysaccharide for a further 24 h before harvesting (“Mature DCs”). The DC phenotype was characterized as CD14−CD11c+CD83+HLA-DRHI by flow cytometry 26-28.
Parenchymal and endothelial cells
Human umbilical vein endothelial cells (HUVECs) (Stem Cell Technologies, Vancouver, BC, Canada) and human renal proximal tubular epithelial cells (RPTECs) (Lonza, Inc., Allendale, NJ) were maintained in tissue culture according to supplier recommendations. Cryopreserved primary RPTECs were purchased from Cambrex (Walkersville, MD) and grown in REGM in 50 mL flasks to 50% confluence. At this point, cells were seeded onto collagen-coated inserts of 6-well plates and grown to 80% confluence in REGM. RPTECs and HUVECs were left untreated or treated with recombinant human IFNG (500 U/mL) for 24 h. After 48 h the cells were collected and stored in TRIZOL at −70°C. Total RNA was extracted and used for analysis by microarrays 29.
Pathway analysis
Transcripts were analyzed using IPA® (Ingenuity® Systems, www.ingenuity.com), with focus on the Canonical Pathways. Significance of enrichment was calculated using Fischer's exact test with Benjamini–Hochberg correction. Pathway diagrams were generated using IPA®.
Results
TCMR classifier transcript comparison reveals a high association of CTLA4 with TCMR
The published TCMR-classifier identified the 30 transcripts most associated with TCMR in a Discovery Set of 403 kidney transplant indication biopsies 18. They were defined by the number of times they were selected in the top 20 TCMR-associated transcripts in 1000 tests of random subsets from the Discovery Set. We studied the robustness of their association with TCMR in the Validation Set of 300 biopsies 19 (Table 2). Every TCMR classifier transcript in the Discovery Set was highly associated with TCMR in the Validation Set, with t-test p-values <10−10.
| Probe set ID | Name | Symbol | Probe set rank (association with TCMR in the Discovery Set (p-value))1 | Corresponding probe set rank in the Validation Set (p-value) |
|---|---|---|---|---|
| 219385_at | SLAM family member 8 | SLAMF8 | 2 (2.2 × 10−27) | 2 (3.7 × 10−20) |
| 235735_at | Tumor necrosis factor (ligand) superfamily, member 8 | TNFSF8 | 3 (3.3 × 10−26) | 520 (4.6 × 10−12) |
| 206761_at | CD96 molecule | CD96 | 4 (1.4 × 10−25) | 145 (1.4 × 10−14) |
| 220485_s_at | Signal-regulatory protein gamma | SIRPG | 5 (3.7 × 10−25) | 4 (9.0 × 10−20) |
| 236226_at | B and T lymphocyte associated | BTLA | 6 (2.3 × 10−24) | 16 (9.4 × 10−19) |
| 210354_at | Interferon, gamma | IFNG | 7 (3.7 × 10−24) | 20 (2.4 × 10−18) |
| 236099_at | – | (Unannotated)2 | 8 (4.9 × 10−24) | 29 (1.3 × 10−17) |
| 203760_s_at | Src-like-adaptor | SLA | 9 (5.3 × 10−24) | 89 (1.6 × 10−15) |
| 239196_at | Ankyrin repeat domain 22 | ANKRD22 | 10 (7.2 × 10−24) | 3 (5.0 × 10−20) |
| 238629_x_at | – | (Unannotated)2 | 14 (1.4 × 10−23) | 18 (1.5 × 10−18) |
| 206545_at | CD28 molecule | CD28 | 18 (6.6 × 10−23) | 300 (3.1 × 10−13) |
| 227458_at | CD274 molecule | CD274 | 23 (1.2 × 10−22) | 27 (7.2 × 10−18) |
| 207777_s_at | SP140 nuclear body protein | SP140 | 25 (3.3 × 10−22) | 294 (2.9 × 10−13) |
| 210116_at | SH2 domain containing 1A | SH2D1A | 33 (7.2 × 10−22) | 279 (2.4 × 10−13) |
| 206134_at | ADAM-like, decysin 1 | ADAMDEC1 | 34 (7.9 × 10−22) | 14 (6.2 × 10−19) |
| 229437_at | MicroRNA 155 | MIR155 | 40 (1.4 × 10−21) | 163 (2.7 × 10−14) |
| 237753_at | Interleukin 21 receptor | IL21R | 42 (2.0 × 10−21) | 81 (1.1 × 10−15) |
| 1552584_at | Interleukin 12 receptor, beta 1 | IL12RB1 | 44 (3.3 × 10−21) | 85 (1.3 × 10−15) |
| 228737_at | TOX high mobility group box family member 2 | TOX2 | 47 (4.2 × 10−21) | 556 (6.0 × 10−12) |
| 206486_at | Lymphocyte-activation gene 3 | LAG3 | 52 (9.6 × 10−21) | 8 (1.4 × 10−19) |
| 204852_s_at | Protein tyrosine phosphatase, nonreceptor type 7 | PTPN7 | 54 (1.1 × 10−20) | 72 (4.9 × 10−16) |
| 215925_s_at | CD72 molecule | CD72 | 60 (1.7 × 10−20) | 10 (2.2 × 10−19) |
| 220423_at | Phospholipase A2, group IID | PLA2G2D | 64 (1.9 × 10−20) | 789 (6.2 × 10−11) |
| 1569225_a_at | Sex comb on midleg-like 4 (Drosophila) | SCML4 | 65 (1.9 × 10−20) | 706 (2.7 × 10−11) |
| 205758_at | CD8a molecule | CD8A | 109 (7.4 × 10−19) | 22 (4.3 × 10−18) |
| 240070_at | T cell immunoreceptor with Ig and ITIM domains | TIGIT | 114 (8.7 × 10−19) | 40 (5.8 × 10−17) |
| 227030_at | IKAROS family zinc finger 3 (Aiolos) | IKZF3 | 123 (1.4 × 10−18) | 444 (1.9 × 10−12) |
| 205242_at | Chemokine (C-X-C motif) ligand 13 | CXCL13 | 196 (2.5 × 10−17) | 372 (8.3 × 10−13) |
| 217147_s_at | T cell receptor associated transmembrane adaptor 1 | TRAT1 | 287 (2.0 × 10−16) | 860 (1.2 × 10−10) |
| 1558972_s_at | Thymocyte selection associated | THEMIS | 301 (2.8 × 10−16) | 518 (4.5 × 10−12) |
- The bolded transcripts are those selected as TCMR classifier transcripts in both the Discovery Set and the Validation Set.
- 1Probe sets are ranked on the association strength, p-value (single pass Bayesian t-test), with T cell–mediated rejection (TCMR) in the Discovery Set (as presented in Figure 1a), or with TCMR in the Validation Set (as presented in Figure 1b).
- 2 At the time of publication, 236099_at and 238629_x_at are poorly annotated Affymetrix probe sets. 238629_x_at is putatively annotated as OR2I2P according to BLAST (http://blast.ncbi.nlm.nih.gov/).
For comparison, we derived 30 TCMR classifier transcripts in the reverse direction, that is, selected in 1000 tests of random subsets from the Validation Set using the same algorithm as was used to define the 30 TCMR classifier transcripts in the Discovery Set. Every Validation Set classifier transcript was highly associated with TCMR in the Discovery Set, with t-test p-values <10−7 (Table S1). The p-values were less significant in the Validation Set because it contained fewer biopsies.
Eleven probe sets were shared between the Discovery Set and Validation Set classifiers. We studied why some Discovery Set TCMR classifier transcripts were not selected in the Validation Set and vice versa. One factor was random variations in rank among highly TCMR-associated transcripts between the Discovery and Validation Sets, which caused them to be missed by the selection algorithm based on top 20 rankings in 1000 tests. For example, the classifier algorithm selected TNFSF8 (CD30 ligand) in all 1000 tests in the Discovery Set 18, but not once in the Validation Set. While TNFSF8 was strongly associated with TCMR by p-value in the Validation Set (p = 4.6 × 10−12), it ranked 520th, precluding selection. Random variations in rank are expected when large numbers of correlated genes are competing for selection 30.
Conversely, CTLA4 was a top transcript in the Validation Set but was not selected in the Discovery Set. This proved to be an inadvertent consequence of interquartile range (IQR) variance filtering, which can miss transcripts with small but highly significant changes. The TCMR classifier transcripts were derived after IQR filtering 18. When we reanalyzed the Discovery Set without IQR filtering, CTLA4 was ranked first among all probe sets (p = 5.9 × 10−28), similar to its number five ranking in the Validation Set (Table S1). Thus variance filtering had deleted the top ranked transcript in the Discovery Set, CTLA4. Subsequent analyses in this study avoided variance filtering.
Global TCMR transcript landscape in the Discovery Set versus the Validation Set
The hierarchy of all transcript changes in TCMR versus all other diseases was conserved between the Discovery Set (Figure 1a) and the Validation Set (Figure 1b). The association strength (t-test p-value) on the x-axis is plotted against the fold change in expression on the y-axis. The TCMR classifier transcripts from the Discovery Set 18 and the position of CTLA4 are shown. We also compared the Discovery and Validation Sets in terms of fold change (Figure 1c) and (p-value) (Figure 1d), with the FDR at 0.05 indicated. In both biopsy sets, fold change (Figure 1c) and p-value (Figure 1d) were strongly conserved across thousands of transcripts.
Note that the p-value for the association with TCMR versus other diseases did not necessarily reflect the fold increase in expression. The highest fold change belonged to CXCL13 and ADAMDEC1, but the most significant association with TCMR was for SLAMF8, CTLA4, SIRPG, ANKRD22, BTLA and IFNG.
When the diagnoses were randomly shuffled for all biopsies as a control, no probe set was associated with TCMR at FDR < 0.05 (data not shown).
The TCMR transcript landscape in the Combined Set of 703 biopsies
- The 30 Discovery Set TCMR classifier transcripts (red).
- B7 family of immune-regulatory (costimulatory and coinhibitory) receptors and their ligands (green).
- Biomarkers of “rejection” 12, including IFNG-inducible chemokines CXCL9, CXCL10 and CXCL11 31, 32 and cytotoxicity-related molecules GZMA, GZMB, PRF1 and GNLY 33-35 (black).
- Inflammasome-related transcripts 8, CASP1, CASP8, IL1B, IL18, PYCARD and AIM2 (beige).
- IFNG-inducible transcripts, annotated in mouse transplants and confirmed in human transplants 36, 37 (yellow).
- Toll-like receptors TLR7, TLR8 and TLR9 (purple).
- AKI-associated transcripts 38 (blue) and kidney solute carriers (pink) 10.
Many members of the B7 family of receptors and their ligands were among the most TCMR-specific transcripts (see Table 3 for details), including CTLA4, BTLA, ICOS, CD274 (PDL1), CD86, and CD28 (PDL2) and CD80 were moderately associated (p-values approximately 10−23), and PD1, CD276 (B7-H3) and VTCN1 (B7-H4) were not (10−5 < p < 10−1). Coinhibitor expression was correlated with interstitial inflammation (i-score): CTLA4, Spearman r = 0.44, p = 2.1 × 10−34; and PDL1 Spearman r = 0.46, p = 1.7 × 10−38, similar to the proinflammatory molecules such as IFNG (Spearman r = 0.47, p = 2.7 × 10−40).
| Probe set ID | Name | Symbol | Alternate designation | TCMR vs. AOB | Probe set signal | ||||
|---|---|---|---|---|---|---|---|---|---|
| Fold change | p-Value | FDR | TCMR | AOB | Normal kidney | ||||
| 236341_at | Cytotoxic T-lymphocyte-associated protein 4 | CTLA4 | CD152 | 2.95 | 3.5 × 10−45 | 9.6 × 10−41 | 83 | 28 | 25 |
| 236226_at | B and T lymphocyte associated | BTLA | 2.80 | 1.6 × 10−41 | 1.8 × 10−37 | 174 | 62 | 47 | |
| 210439_at | Inducible T-cell co-stimulator | ICOS | CD278 | 2.22 | 3.6 × 10−39 | 2.2 × 10−35 | 102 | 46 | 42 |
| 227458_at | CD274 molecule | CD274 | PD-L1, B7-H1 | 2.81 | 1.0 × 10−37 | 4.0 × 10−34 | 427 | 152 | 87 |
| 205686_s_at | CD86 molecule | CD86 | B7-2 | 2.46 | 8.1 × 10−35 | 1.4 × 10−31 | 183 | 74 | 53 |
| 206545_at | CD28 molecule | CD28 | 2.78 | 1.9 × 10−33 | 1.8 × 10−30 | 154 | 55 | 46 | |
| 1554519_at | CD80 molecule | CD80 | B7-1 | 1.54 | 4.5 × 10−24 | 4.8 × 10−22 | 68 | 44 | 43 |
| 220049_s_at | Programmed cell death 1 ligand 2 | PDCD1LG2 | PD-L2, CD273, B7-DC | 1.48 | 2.4 × 10−23 | 2.3 × 10−21 | 47 | 32 | 27 |
| 228976_at | Inducible T-cell co-stimulator ligand | ICOSLG | CD275, B7-H2 | 0.89 | 3.2 × 10−5 | 2.1 × 10−4 | 146 | 164 | 188 |
| 224859_at | CD276 molecule | CD276 | B7-H3 | 1.05 | 8.9 × 10−2 | 1.6 × 10−1 | 407 | 387 | 326 |
| 207634_at | Programmed cell death 1 | PDCD1 | PD1, CD279 | 1.03 | 3.2 × 10−1 | 4.3 × 10−1 | 177 | 172 | 192 |
| 219768_at | V-set domain containing T cell activation inhibitor 1 | VTCN1 | B7-H4 | 1.00 | 9.6 × 10−1 | 9.7 × 10−1 | 392 | 390 | 322 |
- The bolded transcripts are annotated as among the top 50 TCMR-associated transcripts.
Previously described biomarkers of rejection (e.g. GZMB and CXCL9), inflammasome transcript CASP1 (note the six CASP1 probe sets, all in the same range) and IFNG-induced transcripts had moderate associations with TCMR (p-values approximately 10−18). The most strongly TCMR-associated inflammasome regulator was AIM2 (p = 1.0 × 10−23) (Figure 2), possibly reflecting induction in macrophages by IFNG—see below. TLR7 and TLR8 were moderately associated with TCMR, whereas TLR9 was not.
Many AKI-associated transcripts were increased and many solute carrier transcripts were decreased in TCMR, reflecting parenchymal injury and dedifferentiation. These associations were weaker because their expression is altered in other types of injury and disease 9, 13.
We examined the TCMR transcript landscape in early and late TCMR and in subgroups of TCMR with various combinations of i-score, t-score and v-score, but found no differences in the top transcripts, only changes in intensity (J. M. Venner, L. G. Hidalgo, K. S. Famulski, J. Chang and P. F. Halloran, manuscript in preparation). Thus our data indicate the TCMR molecular signals are stereotyped, regardless of details of histologic lesions or time of presentation.
Top TCMR-associated transcript expression in cultured human cells
- effector T cells, e.g. CTLA4, CD28, ICOS, SIRPG;
- effector T cells shared with NK cells, e.g. IFNG, CD96, SH2D1A;
- APCs, including:
- macrophages and DCs, e.g. ADAMDEC1, SLAMF8, CD274 (PDL1); PDL2;
- monocytes, e.g. TNFSF8;
- B cells, e.g. BTLA, CD72;
- IFNG-treated macrophages, e.g. ANKRD22, AIM2.
The top TCMR-associated transcripts were poorly expressed in normal kidney and in RPTECs and HUVECs without IFNG.
Pathway analysis of the TCMR-associated transcripts
We examined the TCMR-associated transcripts for representation in the curated IPA® Canonical Pathways. They were most strongly represented in three pathways: CTLA4 signaling (adjusted p = 1.9 × 10−7), T cell receptor signaling (adjusted p = 1.9 × 10−7) and T helper cell differentiation (adjusted p = 5.5 × 10−7) (Table 4). As shown in the pathway illustration (Figure S1), there was extensive sharing of transcripts among these pathways, which were related to surface membrane molecules of T cell activation signal 1 (CD3D, CD8A, CD8B) and signal 2 (CD28, CD86, ICOS, CTLA4); intracellular signaling LCP2 (SLP-76) and PTPN7 (tyrosine phosphatase); and IL12RB1, IL21R, and IFNG.
| Ingenuity Canonical Pathway (no. of members in pathway) | No. of TCMR-associated transcripts in pathway | Benjamini–Hochberg adjusted p-value | Transcripts |
|---|---|---|---|
| CTLA4 signaling in cytotoxic T lymphocytes (96) | 7 | 1.9 × 10−7 | CD28, CD86, CD8A, CD3D, CTLA4, CD8B, LCP2 |
| T cell receptor signaling (109) | 7 | 1.9 × 10−7 | CD28, PTPN7, CD8A, CD3D, CTLA4, CD8B, LCP2 |
| T helper cell differentiation (72) | 6 | 5.5 × 10−7 | IFNG, CD28, IL12RB1, IL21R, ICOS, CD86 |
| Communication between innate and adaptive immune cells (112) | 5 | 6.6 × 10−5 | IFNG, CD28, CD86, CD8A, CD8B |
| Primary immunodeficiency signaling (64) | 4 | 1.4 × 10−4 | ICOS, CD8A, TAP1, CD3D |
| CD28 signaling in T helper cells (136) | 5 | 1.4 × 10−4 | CD28, CD86, CD3D, CTLA4, LCP2 |
| Role of NFAT in regulation of the immune response (200) | 5 | 8.1 × 10−4 | CD28, CD86, CD3D, LCP2, FCGR1B |
| iCOS-iCOSL signaling in T helper cells (126) | 4 | 1.4 × 10−3 | CD28, ICOS, CD3D, LCP2 |
| Type I diabetes mellitus signaling (121) | 4 | 1.4 × 10−3 | IFNG, CD28, CD86, CD3D |
| PKCθ signaling in T lymphocytes (144) | 4 | 1.5 × 10−3 | CD28, CD86, CD3D, LCP2 |
| Graft-versus-host disease signaling (51) | 3 | 1.5 × 10−3 | IFNG, CD28, CD86 |
| Systemic lupus erythematosus signaling (256) | 5 | 1.6 × 10−3 | CD28, CD72, CD86, CD3D, FCGR1B |
| Hematopoiesis from pluripotent stem cells (63) | 3 | 1.7 × 10−3 | CD8A, CD3D, CD8B |
| Nur77 signaling in T lymphocytes (64) | 3 | 2.0 × 10−3 | CD28, CD86, CD3D |
| Allograft rejection signaling (97) | 3 | 6.3 × 10−3 | IFNG, CD28, CD86 |
| Altered T cell and B cell signaling in rheumatoid arthritis (100) | 3 | 6.5 × 10−3 | IFNG, CD28, CD86 |
| Crosstalk between dendritic cells and natural killer cells (106) | 3 | 6.9 × 10−3 | IFNG, CD28, CD86 |
| Natural killer cell signaling (118) | 3 | 1.2 × 10−2 | SH2D1A, LAIR1, LCP2 |
| Interferon signaling (36) | 2 | 1.3 × 10−2 | IFNG, TAP1 |
| Antigen presentation pathway (42) | 2 | 1.5 × 10−2 | IFNG, TAP1 |
| Autoimmune thyroid disease signaling (62) | 2 | 2.3 × 10−2 | CD28, CD86 |
| Role of MAPK signaling in the pathogenesis of influenza (72) | 2 | 4.5 × 10−2 | IFNG, RARRES3 |
We also examined the top 300 TCMR-associated transcripts in Canonical Pathways. The same three pathways were confirmed as being over represented: T cell receptor signaling (adjusted p = 6.7 × 10−13), CTLA4 signaling (adjusted p = 1.1 × 10−12) and T helper cell differentiation (adjusted p = 1.3 × 10−9). This analysis implicated many T cell activation molecules, as indicated in Figure S1.
We conclude that the simplest explanation for over representation of the top TCMR-associated transcripts in Canonical pathways was T cell receptor triggering in effector T cells leading to IFNG expression, with an unexpected prominence of costimulatory/coinhibitory signaling between the effector T cell and APCs.
Discussion
We characterized the transcript expression landscape in pure TCMR versus other diseases in two consecutive but independent populations of indication biopsies, a Discovery Set and a Validation Set. This study was possible because of the development of conventional (i.e. histologic) diagnostic criteria to distinguish pure TCMR from ABMR, including C4d-negative ABMR, preventing contamination of TCMR biopsies with undetected ABMR and providing a critical comparator to distinguish events related to cognate recognition from those related to the inflammatory response. The fold changes and association strength (p-values) for thousands of transcripts were strongly conserved in the two independent populations, indicating a strict hierarchy. The top transcripts were expressed in effector T cells (some shared with NK cells); in APCs (DCs, macrophages and B cells); or were IFNG-inducible in macrophages. An unexpected finding was the prominence of many costimulation-related transcripts such as CTLA4 and PD1 ligands. Many reported rejection biomarkers had significant but weaker associations, presumably reflecting associations with inflammation (e.g. CXCL9, IFNG effects and inflammasome-related transcripts CASP1, CD95, IL1B and IL18), and parenchymal responses to injury (injury-induced transcripts and loss of kidney solute carriers). In pathway analysis, the top TCMR-associated transcripts reflected proximal pathways of T cell–APC interactions, particularly T cell receptor triggering and costimulation through B7 family members and their ligands. We conclude that the massive disturbance in gene expression in TCMR is highly structured, with the signal that is most characteristic of TCMR versus other diseases is the expression of molecules reflecting a cognate engagement between effector T cells and APCs, featuring T cell receptor triggering and positive and negative costimulation.
We postulate that the top TCMR-associated transcripts reflect transcriptional regulation in T cells and/or APCs as a consequence of the immunologic synapse 39, 40. While existing in vitro systems cannot model the full complexity of effector T cell–APC interactions in rejecting tissues, many molecules reported here are increased in models of T cell–APC synapses. In T cells, these include proximal receptor and co-receptor signaling systems such as CD3D, CD3Z (CD247), CD8A, CD8B 41, ICOS 42, CTLA4 42, CD96 42, SIRPG 43, SH2D1A 43, phosphatases PTPN6 (SHP1) 42 and PTPN7 42, TRAT1 (a CTLA4 chaperone 44); kinases LCK 43 and ITK 43 and IFNG itself 42, 45, all identified as TCMR classifier or TCMR-associated transcripts in the present study.
We believe that IFNG emerged as the principal cytokine in all TCMR classifier and TCMR-associated transcript lists in both the Discovery and Validation Sets because IFNG transcription indicates antigen-specific triggering of primed effector T cells 45. Moreover, the fact that two highly ranked transcripts—AIM2, a key inflammasome regulator, and ANKRD22—were IFNG-inducible in macrophages, while most IFNG-inducible genes are less strongly associated with TCMR, raises the possibility that these IFNG effects have a special relationship to the cognate synapse, as opposed to more general effects of IFNG.
The striking association of pure TCMR with increased expression of many B7 family members, including CTLA4, PDL1, PDL2, ICOS and CD28, suggests unexpected roles for these molecules in regulating the effector T cell–APC interactions within the target tissue, beyond their role in controlling T cell triggering and effector T cell generation in the secondary lymphoid organs. Given the low requirement for costimulation in triggering effector T cells 46, 47, the main role of B7 family members and their ligands within inflammatory compartments may be in negative regulation of effector T cells, as has been previously implicated for CTLA4 in rejecting human kidney allografts 48. This could be a factor in the unexplained occurrence of vigorous TCMR in patients immunosuppressed with belatacept (CTLA4Ig), which blocks CD80 and CD86 49. If CTLA4, and potentially other regulators, exert active negative regulation of effector T cells in tissues with TCMR 49, blocking CD80 and CD86 with belatacept could release excessive effector T cell activity, more than counteracting the immunosuppressive effect on effector T cell generation. The strong association of PDL1 and PDL2 with TCMR and their importance in regulating immunity and tolerance also invites consideration of their roles in regulating T cells in the inflammatory compartment 50, 51. This is relevant to the recent successes in inducing a TCMR-like process in cancer via interventions against CTLA4 and PD1 1: raising the possibility that targeting these inhibitory elements could amplify effector T cell activity in cancer tissue.
The macrophage transcript ADAMDEC1 (decysin-1), which encodes a secreted metzincin metalloprotease 52, was strongly TCMR-associated and highly expressed in macrophages in our culture conditions. ADAMDEC1 is also expressed in germinal center DCs 53, but whether the myeloid cells expressing ADAMDEC1 in TCMR are best defined as macrophages, DCs or intermediate differentiation states cannot not be determined because markers reliably distinguishing these cells in inflamed tissues are lacking. Of interest, ADAMDEC1 is highly expressed in sarcoidosis, another potential cognate T cell–mediated process 54.
The prominence of CD72, a B cell ligand for CD100 and CD5 on T cells implicate B cells as APCs in sustaining cognate recognition within the tissue. B cells are an antigen-specific APC and can clonally expand and mature, giving them advantages in sustaining T cell inflammatory responses.
While some of the top TCMR-associated transcripts were expressed in both NK cells and effector T cells, their expression in TCMR is best explained by effector T and NKT cells rather than NK cells because no transcripts were exclusive for NK cells. NK cells are difficult to study in tissues because of the paucity of markers that distinguish NK cells from NKT and effector T cells, but previous studies suggest they are rare 23, 25, 55, 56. NK transcripts are found in ABMR 25, 29, but these are distinct from those in TCMR and are not accompanied by T cell transcripts.
The strong conservation of the molecular landscape of TCMR in both biopsy sets—the hierarchy of molecular associations and fold changes—refutes the criticism that molecular phenotyping is poorly reproducible between biopsy sets 57. Variations in rank are expected between high dimensionality data sets with similar biology, because of random variation among tens of thousands of probe sets. Focusing on differences in the “top transcripts” can be misleading because of predictable random variation, which can miss the remarkable stereotyping of thousands of molecular changes.
The present study, like recent studies in cancers 58, illustrates the potential utility of analyzing sets of thoroughly phenotyped tissues as models of human diseases, combined with insights from in vitro and animal models (reviewed in 59, 60). In vitro models, even three-dimensional models, cannot simulate diseased tissues because they lack influences such as transendothelial migration, extracellular matrix interactions and systemic influences. Immunostaining is a critical tool but is limited by sampling error, uncertainty about cross-reactivity and a paucity of reagents for many proteins of interest. Small animal models of kidney transplant rejection do not manifest the pathology of human ABMR, a key comparator in the present study, and are not subject to influences such as immunosuppression that may modify the human diseases states. Cell isolation from needle biopsies is limited by biopsy size. We believe that high dimensionality molecular analysis of well phenotyped human biopsy populations are emerging as a promising strategy for the basic understanding of human diseases.
Acknowledgments
This research has been supported by funding and/or resources from Novartis Pharma AG, and in the past by Genome Canada, the University of Alberta Hospital Foundation, Roche Molecular Systems, Hoffmann-La Roche Canada Ltd., Canada Foundation for Innovation, the Alberta Ministry of Advanced Education and Technology, the Roche Organ Transplant Research Foundation, and Astellas. Dr. Halloran held a Canada Research Chair in Transplant Immunology until 2008 and currently holds the Muttart Chair in Clinical Immunology.
Disclosure
The authors of this manuscript have conflicts of interest to disclose as described by the American Journal of Transplantation. P. F. Halloran holds shares in Transcriptome Sciences, Inc., a company with an interest in molecular diagnostics. The other authors have no conflicts of interest.
