Professional Documents
Culture Documents
DOI: 10.1111/ajt.14600
MINIREVIEW
1
Alberta Transplant Applied Genomics Centre,
Edmonton, AB, Canada The molecular mechanisms operating in human organ transplant rejection are best in-
2
Department of Medicine, Division ferred from the mRNAs expressed in biopsies because the corresponding proteins
of Nephrology and Transplant often have low expression and short half-lives, while small non-coding RNAs lack
Immunology, University of Alberta, Edmonton,
AB, Canada specificity. Associations should be characterized in a population that rigorously identi-
3
Medizinische Hochschule Hannover, fies T cell-mediated (TCMR) and antibody-mediated rejection (ABMR). This is best
Hannover, Germany
achieved in kidney transplant biopsies, but the results are generalizable to heart, lung,
4
Department of Laboratory Medicine and
Pathology, University of Alberta, Edmonton,
or liver transplants. Associations can be universal (all rejection), TCMR-selective, or
AB, Canada ABMR-selective, with universal being strongest and ABMR-selective weakest. Top
Correspondence
universal transcripts are IFNG-inducible (eg, CXCL11 IDO1, WARS) or shared by ef-
Philip F. Halloran fector T cells (ETCs) and NK cells (eg, KLRD1, CCL4). TCMR-selective transcripts are
Email: phallora@ualberta.ca
expressed in activated ETCs (eg, CTLA4, IFNG), activated (eg, ADAMDEC1), or IFNG-
induced macrophages (eg, ANKRD22). ABMR-selective transcripts are expressed in
NK cells (eg, FGFBP2, GNLY) and endothelial cells (eg, ROBO4, DARC). Transcript as-
sociations are highly reproducible between biopsy sets when the same rejection defi-
nitions, case mix, algorithm, and technology are applied, but exact ranks will vary.
Previously published rejection-associated transcripts resemble universal and TCMR-
selective transcripts due to incomplete representation of ABMR. Rejection-associated
transcripts are never completely rejection-specific because they are shared with the
stereotyped response-to-injury and innate immunity
KEYWORDS
basic (laboratory) research/science, biopsy, kidney transplantation/nephrology, organ
transplantation in general, rejection
1 | INTRODUCTION how molecular measurements are used for diagnosis, which is covered
elsewhere.1 The main focus is on kidney transplant biopsies because
This minireview discusses the protein-coding transcripts (mRNAs) as- the data sets are large and the rejection states are better character-
sociated with rejection in transplant biopsies and the insight they pro- ized (by histology and by mRNA expression) than for other transplants
vide into mechanisms and disease classification. We will not discuss (as indicated by reproducibility of TCMR between pathologists: kidney
50%2; heart 28%3; and lung 18%4). In this review, we also compare
our findings in the Molecular Microscope® project1 to other published
Abbreviations: ABMR, antibody-mediated rejection; AKI, acute kidney injury; ARTs, acute
rejection transcripts; AvEET, ABMR vs. everything else plus TCMR; cAR, clinical acute rejec- literature, taking into account that many published biopsy studies used
tion; CRM, common rejection module; ETCs, effector T cells; ICR, immunological constant of different or outdated histology classifications and did not include or
rejection; RTW, response to wounding; RvEEABMR, TCMR and mixed ABMR/TCMR rejec-
tion vs. everything else; TCMR, T cell-mediated rejection; TvEEA, TCMR vs. everything else
recognize ABMR. Details are provided in recent reviews1 and in our
plus ABMR. recent analysis of universal and selective rejection associations.5
Changes in mRNA expression that are associated with rejection for research and discovery as the alternative mRNA splicing and al-
present opportunities to infer mechanisms, identify druggable targets, ternative promoters add intriguing levels of complexity and potential
develop new diagnostic tests, and correct errors in the empirical his- specificity that have not yet been explored. An alternative to RNAseq
tology system.6,7 TCMR itself is also a model system for understand- is the new Clariom D microarray, which can identify >540 000 coding
ing human T cell-mediated inflammatory diseases such as ulcerative and non-coding transcripts.
colitis and multiple sclerosis, and the mechanisms of tumor rejection The identity of the mRNAs associated with rejection is inferred
activated by checkpoint inhibitors.8 Associations are only one pathway from the microarray probe set with which it hybridizes. This distinc-
in developing evidence but they are an important first step in under- tion is important when considering mRNA that may have been chem-
standing diseases, just as associated findings are important clues in a ically modified by formalin-fixation and paraffin-embedding (FFPE).
crime scene investigation. Formalin chemically modifies mRNA and produces both false positive
Transcripts increased or decreased in biopsies with rejection may and false negative hybridization.15,16 The use of FFPE tissue for di-
reflect transcription changes in resident cells (eg, IFNG-inducible agnostic assessment is being examined in some centers but it is not
changes in the donor endothelium) or changes in cell populations, eg, the preferred method for studies discovering transcript associations,
infiltration and activation of ETCs and macrophages in TCMR or mar- which typically require biopsies stabilized in RNAlater or snap-frozen
gination and activation of intracapillary NK cells in ABMR. Therefore to avoid chemical modification.
we never use the terms “upregulated” and “downregulated,” as “in- The raw data from microarrays requires normalization of the data-
creased” or “decreased” is both more accurate and more descriptive. set, and in many cases the data is variance filtered (ie, probe sets with
Interpretation of transcript changes can be assisted by studying rele- low variance are removed) to reduce the “noise,” the potential contam-
vant mouse models, cell lines, and the literature for each gene (https:// ination by accidental differences. The risk in variance filtering is that
www.ncbi.nlm.nih.gov/gene). important low-expression signals may be discarded. For example, in
The identification of the mRNAs associated with a conventionally studying TCMR, variance filtering initially missed CTLA, a transcript
defined disease state depends on the accuracy of the designation of with low signal strength but one of the most highly associated with
the disease state. The true disease state always remains a “latent vari- TCMR.2,17,18
able”—one that is not directly observed but rather inferred by rules or We use large data sets that include many well-defined TCMR
algorithms from other variables that are directly measured. cases, both early and late-stage ABMR cases (including C4d neg-
ative ABMR); and many other diseases and injuries. Identifying
TCMR- or ABMR-selective transcripts requires inclusion of the op-
1.1 | Why study mRNA?
posite phenotype in the comparator.19 We generally study biopsies
Transcripts encoding proteins can be amplified whereas their protein for clinical indications (eg, dysfunction, proteinuria, investigation
products are often undetectable. Small non-coding RNAs (eg, miRNAs of DSA) in consented patients in prospective trials (Clinicaltrials.
and siRNA) provide less biological insight than mRNAs because they gov NCT01299168) because rejection phenotypes are much more
have less diversity by an order of magnitude and lack specificity for frequent and severe in indication biopsies than protocol biopsies
individual genes due to their promiscuous regulation of immune and of stable patients, though conclusions are applicable to protocol
non-immune processes. To date, small non-coding RNAs have shown biopsies.20
9-12
limited ability to define rejection classes.
The top ranked rejection-associated transcripts in one study pop- the actual phenotype of each disease manifests the selective plus the
ulation will not necessarily predict the top-ranked transcripts in a new universal features. Thus, TCMR or ABMR can be annotated by their
population because rank is influenced by random error (“noise”) as well selective features and rejection transcripts (Table 1).
as the prevalence of disease states (“signal”). The problem should be
considered whenever any “top gene” list is presented: the goal is to
3.1 | Universal rejection-associated transcripts
discover strong associations with different disease states, but the top-
ranked transcripts are only a sketch. Generally, the strong associations This list is dominated by transcripts that are IFNG-inducible in pa-
of transcripts with a disease state are preserved when exact rank is renchymal cells, endothelial cells, and/or in macrophages eg, CXCL9,
not because they reflect specific biologic mechanisms operating in all WARS, IDO, and HLA-E. In addition, about 10% are shared by acti-
similar study cohorts. vated ETCs and NK cells, eg, KLRD1, CCL4 and PRF1. IFNG inducibil-
ity is the principal feature of the universal transcripts and IFNG itself
ranks at the top of the transcripts induced in activated ETCs and NK
2.3 | Distinguishing rejection associations from
cells.29 The ranks of top 100 universal rejection transcripts were high
response-to-wounding (RTW)
in both ABMR-selective and TCMR-selective transcript lists. Pathway
Tissue injury sustained from donation and implantation, rejection, and analyses confirmed the prominence of IFNG-STAT mechanisms.5
diseases affects all transplanted organs, even live-donor organs, and
it evokes a RTW. New diseases or recurrence of primary disease also
3.2 | TCMR-selective transcripts
induce the RTW. Thus the RTW (recently reviewed22) impacts expres-
sion of thousands of transcripts that are shared across various injury The top TCMR-selective transcripts reflect effector T cell and mac-
states and disease (including rejection), and involves parenchymal rophage infiltration and activation. Effector T cell transcripts in-
cells, stroma, and microcirculation endothelium, as well as recruitment clude IFNG; checkpoint molecules CTLA4, ICOS, and BTLA; CD96,
and activation of inflammatory cells. RTW is underestimated by his- LAG3, SIRPG, LCP2, DUSP2, and CD8A. ANKRD22 (IFNG-inducible
tology and evolves over months to years.23,24 Distinguishing rejection in macrophages) and non-IFNG-inducible macrophage molecule
from response-to-wounding (RTW) requires comparison of the posi- ADAMDEC1 are highly TCMR-selective. TCMR-selective transcripts
tive class to a population that includes a wide variety of non-rejection rank high among the universal rejection transcripts (mean 596) but
disease states, as well as relatively normal tissue (“everything else”). low in ABMR-selective transcripts (mean 33573).2,18,30 In pathway
analysis of TCMR-selective transcripts, the top pathways included
T cell receptor signaling and CTLA4 costimulation. Other features of
3 | REJECTION-A SSOCIATED myeloid cells include CD80 and CD86, CD274/PD-1 ligand 1; CD84,
TRANSCRIPTS: UNIVERSAL, TCMR- IL21R, LAIR1, PHEX, SLA, SLAMF8, TNFSF8/CD30.
SELECTIVE, AND ABMR-S ELECTIVE
3.3 | ABMR-selective transcripts
We used a discovery set—validation set approach of prospectively
collected biopsies to document the rejection-associated transcript The ABMR-selective transcripts (eg, CCL4, CXCL11, CXCL10, and
changes that are universal,5,25 TCMR-selective,2,26 and ABMR- PLA1A) rank high among the universal rejection transcripts, but low
selective.17,27 The initial discovery sets had 403 biopsies and valida- in TCMR-selective transcripts.17,27,28 We believe that the ABMR-
tion sets contained 300 independent biopsies. selective transcripts have weaker associations (P-values) than the
Associations were visualized as “landscapes” and top associated TCMR-selective transcripts because the ABMR process is largely con-
transcripts (ie, on the x axis) were used for pathway analyses.5,18,28 fined to the capillary compartment and affects fewer cells, as opposed
(We use the term landscape to refer to volcano plots that visualize to the TCMR process in the expanded interstitium. ABMR-selective
association strength on the x axis and fold change in expression on transcripts are expressed in NK cells (SH2D1B, GNLY, FGFBP2,
the y axis.) Altered expression of thousands of transcripts—fold change CD160), endothelial cells (DARC, ROBO4, CDH5, CDH13), or are
and association strength—was highly conserved between the discov- IFNG-inducible in endothelium (eg, PLA1A and CXCL11).
ery and validation sets. Some ABMR-selective transcripts are decreased (eg, scle-
When we annotated the top transcripts associated with kidney rostin [SOST] and corticotropin releasing hormone binding
27
transplant rejection by t-tests in 703 biopsies, the adjusted P-values protein [CRHBP] ). We postulate that the endothelial injury and de-
show a hierarchy from universal (eg, 10−52 to 10−40) to TCMR-selective differentiation that is not visible histologically or by electron micros-
(10−41 to 10−31) to ABMR-selective (10−20 to 10−9). Thus the associ- copy. These decreases are strongly associated with late-stage ABMR,
ations are strongest for universal changes and weakest for ABMR- but it is unclear how they relate to the histologic changes of late-stage
selective changes. ABMR (eg, capillary basement membrane multi-layering or glomerular
The published landscapes of the transcript changes in the univer- double contours).
sal, ABMR-selective, and TCMR-selective binary algorithms are repro- Pathway analysis of ABMR-selective transcripts identified
duced in the volcano plots in Figure 1. It is important to recognize that angiogenesis-like changes with roles for angiopoietin and vascular
|
788 HALLORAN et al.
F I G U R E 1 Molecular landscapes of TCMR (A), ABMR (B), and all rejection (C). Color keys (right lower quadrant) were assigned to the TCMR-
selective, ABMR-selective, and rejection-associated transcripts (probe sets) as well as to the transcripts representing IFNG effects, B7 family
ligands and receptors, inflammasome, response to wounding, and parenchymal function
endothelial growth factors and leukocyte-endothelial interactions, and probably similar to those in the T cell receptor-activated effector T cell
with evidence for NK CD16a Fc receptor signaling.1,29,31 in the interstitium, eg, IFNG production and CD160 expression.31 In
ABMR-associated transcripts in heart transplant endomyocar- contrast to TCMR, ABMR initially induces more endothelial injury but
dial biopsies are very similar to a mix of the universal transcripts and little parenchymal injury (which is why ABMR is often relatively silent
ABMR-selective transcripts (eg, CXCL11, ROBO4) in kidney trans- clinical problem).
plants with ABMR.32,33
These results suggest models for TCMR and ABMR,1,34 summa-
3.4 | Comparing algorithms (Figure 2)
rized here to encourage experimental validation. In TCMR, the cognate
effector T cells engage antigen on endothelium or antigen present- We compared the shared top transcripts for eight class comparisons
ing cells (APC) prior to transendothelial migration35,36 and cross the to interrogate microarray results from 703 kidney biopsies, 205 with
microcirculation without seriously damaging it. ETCs then engage the rejection. The positive comparators were all rejection, TCMR, or
APCs in the parenchyma and create an inflammatory compartment ABMR. The negative comparators included all non-rejecting biop-
dominated by the cognate engagement of T cell receptor and costim- sies (which include normal biopsies and other transplant diseases).
ulators with APCs (macrophages), regulated by inhibitors like CTLA4, Selectivity for ABMR or TCMR (ie, generation of the ABMR- and
PD-1, ICOS, and BTLA The inflammatory reaction in TCMR always TCMR-selective transcripts) required the other rejection class as well
triggers a RTW in the parenchyma, analogous to acute kidney injury as non-rejection biopsies in the comparator to avoid selecting univer-
(AKI). In ABMR, the NK cell CD16a Fc receptors (FCGR3A) engage sal rejection transcripts.
the Fc portions of IgG bound to HLA on the endothelium, activating When we generate subsets from our biopsy collection, the strong
release of IFNG and other cytokines, potentially releasing cytotoxic associations are always conserved in the class comparisons but the
enzymes, and evoking an injury response in the endothelium. The mo- exact rank shows random variation. In one half of our 703 biopsy set,
lecular events in the CD16a-activated NK cell inside the capillary are the top 100 universal rejection transcripts in each algorithm will be
HALLORAN et al. |
789
T A B L E 1 Most robust associations: Comparing universal (Ref. 5), ABMR-selective (Ref. 27), and TCMR-selective (Ref.17) top 30 transcripts*
RvEE, rejection vs. everything else; AvEET, ABMR vs. everything else including TCMR; TvEEA, TCMR vs. everything else including ABMR; ABMR, antibody-
mediated rejection; TCMR, T cell-mediated rejection.
*Published table of top 30 transcripts in 703 biopsies assessed with U133 microarrays (Ref. 5).
a
Annotation by highest expression in a cell panel, supplemented by literature. Genes that overlap between conditions have underlined symbols. Transcripts
expressed by T cells and NK cells are assigned as NK transcripts in ABMR and T cell transcripts in TCMR. Transcripts expressed in T cells and myeloid cells
are considered T cell transcripts for this summary.
b
Induced by anti CD16a in cultured NK cells.
c
As noted in text, CCL4 is expressed in various cell types and inducible by IFNG treatment. CCL4 is expressed in T and NK cells and inducible in macro-
phages but did not qualify as GRIT by our definition.
only 50-60% shared with the top 100 in the other half, yet all retain After four months post-transplantation, kidney transplants often
strong statistical associations and represent the same biological pro- manifest immunoglobulin transcripts (representing plasma cells)38 and
cess. Again, this illustrates the role of noise in determining exact rank. certain mast cell transcripts.39 These emerge in kidneys developing
atrophy-fibrosis as part of the RTW, and should not be mistaken for
adaptive immune mechanisms.24
3.5 | Rejection-associated transcripts are expressed
during injury-induced RTW
Transcripts associated with rejection are never truly specific for re- 4 | COMPARISON TO PUBLISHED
jection. Many top transcripts in ABMR, TCMR, and all rejection are LISTS OF ACUTE REJECTION-
higher in abnormal non-rejecting than normal transplants (Figure 3). ASSOCIATED TRANSCRIPTS
ABMR-selective and endothelial transcripts are least specific since
they are high in TCMR and non-rejection injuries. This is not surpris- Other published studies of rejection-associated transcripts40–46 differ
ing: the canonical histologic lesions of ABMR (ptc-, g-, cg- lesions) and in platforms, histologic classifications, case mix, and organs studied,
TCMR (t-, i-, and v-lesions) are also not specific, occurring in other which complicates direct comparisons. We have related our rejection
diseases and injury.37 transcripts to lists of transcripts associated with “acute rejection”:
790 | HALLORAN et al.
ABMR = RvEE plus AvEET RvEE 100 TCMR = RvEE plus TvEEA
AvN 76
AvEE 70
TvN 45
AvEET 39 TvEE 37
TvEEA 9
TvA 0
Relatively selective for ABMR Relatively selective for TCMR
Degree of sharing vs. selectivity: based on sharing transcripts associated with all rejection
F I G U R E 2 Illustration of the effect of changing the case mix in the positive and negative comparators on transcript sharing by ten alternative
rejection-associated algorithms. Numbers indicate the non-redundant and annotated transcripts shared with rejection vs. everything else (RvEE)
algorithm, based on the top 100 rejection-associated transcripts. TvN - TCMR vs. Nephrectomies, TvEE - TCMR vs. everything else (excluding
ABMR), TvEET - TCMR vs. everything else including ABMR, TvA - TCMR vs. ABMR, AvEET - ABMR vs. everything else including TCMR, AvEE -
ABMR vs. everything else (excluding TCMR), AvN - ABMR vs. Nephrectomies
ANOVA p<0.001
3.0
2.5
Mean expression of ABMR genes
(fold change vs. normal kidneys)
2.0
liver acute rejection47; Common Rejection Module (CRM)43; Acute rejection transcripts and TCMR-selective transcripts algorithms.5 Our
48
Rejection Transcripts in kidney (ARTs) ; and clinical Acute Rejection approach was to relate these acute rejection transcript lists (rank and
(cAR) in kidney.40 The CRM and the Immunologic Constant of P-values) to our universal rejection transcripts, TCMR-selective, and
49,50
Rejection (ICR) were derived from multiple published lists and ABMR-selective transcripts.5
across organs. All reflect class comparisons of transcript changes in The acute rejection transcripts in liver, CRM, ARTs, and cAR all
biopsies with acute rejection versus non-rejection. The histologic were highly enriched in universal rejection transcripts, eg, CXCL9,
classification of rejection in these studies is outdated, and the case CXCL10, PSMB9, TAP, UBD, and HLA. For example, 46 of 67 ARTs rank
mix is not clear: most biopsies were early when ABMR is infrequent in top 300 universal rejection transcripts, with 26 annotated as IFNG-
(in patients transplanted with no DSA), and were classified without inducible (Table 2). The median rank was lower in TCMR (Table 2)
current criteria for ABMR. From the above consideration, these lists because TCMR selectivity requires ABMR to be annotated and well
should primarily detect the transcripts characterized in our universal represented in the comparator.
T A B L E 2 Comparison of published Acute Rejection transcripts to the RvEE transcripts ordered by their rank in RvEE comparisona
Gene symbol Rank in RvEEb Gene symbol Rank in RvEE Gene symbol Rank in RvEE Gene symbol Rank in RvEE Gene symbol Rank in RvEE
Underlined are transcripts overlapping with the union of rejection transcripts in kidneys (Ref. 5).
RvEE, rejection vs. everything else, TvEEA, TCMR vs. everything else including ABMR, AvEET, ABMR vs. everything else including TCMR.
a
Genes in the Acute Rejection (AR) data sets were selected by comparing biopsies with AR to biopsies from well-functioning transplants.
b
Order by their rank in the RvEE landscape by corrected P-value.
c
Bolded gene symbols indicate IFNG-inducible (http://atagc.med.ualberta.ca/Research/GeneLists/Pages/default.aspx).
|
d
CCL5 expression is inducible by IFNG but is also highly expressed in effector T cells.
e
791
T A B L E 3 Ranking of the top 50 increased transcripts (out of 2653) from the cAR (internal) transcript list (Ref. 40) in the RvEE and TvEEA
comparisons ordered by their rank in the RvEE algorithm
cAR internal
Gene symbol Rank in cAR Rank in RvEEb Rank in TvEEAc Gene symbol Rank in cAR Rank in RvEE Rank in TvEEA
The top 50 cAR transcripts (by P-value) in the original supple- Interestingly, the cAR and ICR lists had similar median P-values for
mentary material40 included our universal rejection transcripts CCL4, universal rejection transcripts and TCMR-selective transcripts, al-
KLRD1, GBP1, and CXCL9 (ranks 2-12) (Table 3). Their median rank in though median ICR P-values were weaker. The median P-values of
our all rejection data set was 522, and in TCMR was 590, indicating ABMR-selective transcripts were poor for all of the acute rejection
universal/selective mix with essentially no ABMR-selectivity. transcript lists. Thus published lists of rejection transcripts resemble
For the ICR,50 of 263 transcripts that mapped to HG 133 2.0 Plus our universal rejection transcripts and TCMR-selective transcripts, yet
Affymetrix arrays, 81 overlapped our top 300 universal rejection lack selectivity for ABMR.
transcripts. The ICR universal rejection transcripts included CXCL11,
CCL4, WARS, CXCL10, and CXCL9 (ranks 1-12) and 49 IFNG-induced
transcripts. 5 | RELEVANCE TO
Overall, these previously published gene lists reflect universal OTHER ORGAN TRANSPLANTS
and TCMR-selective transcripts, (Table 4). The liver acute rejection,
CRM, and ARTs lists had stronger associations with universal rejection The mRNAs associated TCMR and ABMR in kidney transplants are
(median P-values) than with TCMR (nine to six orders of magnitude). highly conserved in heart transplants,32,33 giving us confidence to use
HALLORAN et al. 793|
T A B L E 4 Median rank in the ATAGC
Acute rejection data RvEE median P TvEEA median P AvEET median P
703 dataset of transcripts annotated in
set value* (range) value* (range) value* (range)
other published acute rejection data sets
Liver acute rejection 1.8 × 10−27 2.5 × 10−21 0.03 (3.0 × 10−09-0.98)
(n = 19) (1.2 × 10−47-0.98) (3.2 × 10−31-0.30)
Common rejection 2.2 × 10−32 1.2 × 10−23 3.8 × 10−03
module (CRM) (1.2 × 10−47- (3.2 × 10−31- (3.0 × 10−09-0.72)
(n = 12) 9.2 × 10−10) 2.6 × 10−12)
ARTs study (n = 67) 2.9 × 10−32 3.0 × 10−23 9.0 × 10−3
(3.4 × 10−36-0.04) (3.1 × 10−25-0.03) (3.0 × 10−9-0.53)
CAR study (internal) 1.4 × 10−23 3.5 × 10−21 0.04 (6.1 × 10−12-0.91)
(n = 50)** (1.3 × 10−36-0.64) (3.09 × 10−41-0.35)
Immunological 5.1 × 10−16 5.3 × 10−15 0.07 (8.0 × 10−19-0.99)
constant of rejection (3.6 × 10−52-0.54) (9.6 × 10−41-1.0)
(n = 263)***
RvEE, Rejection versus everything else (non rejecting); TvEEA, TCMR vs. everything else (including
ABMR); AvEET, ABMR vs. everything else (including TCMR).
*Association with rejection (P values with Benjamini-Hochberg correction) in RvEE comparison.
**The top 50 transcripts selected by P values from the original publication were used.
***number of transcripts that mapped to HG133 2.0 Plus Affymetrix arrays.
Three main “universal” (RvEE) most significant P values; IFNG effects and shared effector T/NK cell molecules
algorithms “TCMR-selective” (TvEEA) activated effector T cell and macrophage transcripts; selected IFNG effects
“ABMR-selective” (AvEET) least significant P values; IFNG effects, endothelial injury response; NK cells
Associations and
selectivity Description Comments
Four factors related algorithm, case mix, histology Strong associations are maintained if the four factors are maintained. Exact rank will
to associations: classification system, and differ in otherwise similar populations due to random variation between popula-
measurement platform tions (noise).
Other associations associations with response to Rejection associations are shared with response to wounding: not specific.
wounding and across organs Considerable conservation of rejection-associated transcripts across organs.
Selectivity Selectivity for ABMR or TCMR If ABMR is not included and well characterized, class comparisons regress toward
requires inclusion of the opposite RvEE. Need selectivity for stages in ABMR: early-stage, fully-developed, late-stage.
class and of EE in the comparator.
ABMR, antibody-mediated rejection; TCMR, T cell-mediated rejection; EE, everything else; RvEE, TCMR and ABMR vs. everything else (non rejecting);
TvEEA, TCMR vs. everything else plus ABMR; AvEET, ABMR vs. everything else plus TCMR.
*
Usually annotated by P value in binary class comparison: positive class vs. negative class. An alternative is to use correlations with molecular scores for each
biopsy.
the kidney results as a basis for defining rejection in other organs. For having large numbers of the opposite rejection phenotype included in
liver TCMR this is also true as we have seen in the liver acute rejec- the negative class and proper histologic annotation, so that TCMR and
47
tion study above. We expect this will be the case in lung transplant ABMR do not contaminate each other. Otherwise, the selectivity for
ABMR and TCMR and for liver transplant ABMR, but the histological ABMR or TCMR decreases when the negative class does not include
rejection phenotypes are so poorly defined in lung (and ABMR in liver) non-rejection biopsies and the opposite rejection class. The biology
that this will take time to develop. represented by the top transcripts changes with case mix in the nega-
tive class: for example, for ABMR-selectivity, endothelial changes only
emerge as top transcripts when TCMR and EE are in the negative class.
6 | PRINCIPLES OF MOLECULAR When rejection transcript measurements are used in molecular di-
ASSOCIATION STUDIES agnostics, they should be incorporated into machine-learning derived
classifiers, which increase their accuracy compared to simple measure-
Some principles emerging from our analyses are listed in Table 5. Ranking ments and cut-offs. Machine learning can assign disease diagnoses
of probe sets varies with the positive class and negative class, ie, case using existing classifications but can also be used to discovery new
mix matters for label classes. Selectivity for TCMR or ABMR requires classifications using clustering methods such as archetype analysis.21
|
794 HALLORAN et al.
33. Loupy A, Duong Van Huyen JP, Hidalgo LG, et al. Gene expression therapeutics for organ transplantation. J Exp Med. 2013;210(11):
profiling for the identification and classification of antibody-mediated 2205‐2221.
heart rejection. Circulation. 2017;135(10):917‐935. 44. Dean PG, Park WD, Cornell LD, Gloor JM, Stegall MD. Intragraft gene
34. Reeve J, Halloran PF, Kaplan B. Common errors in the imple- expression in positive crossmatch kidney allografts: ongoing inflam-
mentation and interpretation of microarray studies. Transplant. mation mediates chronic antibody-mediated injury. Am J Transplant.
2015;99(3):470‐475. 2012;12(6):1551‐1563.
35. Walch JM, Zeng Q, Li Q, et al. Cognate antigen directs CD8 + 45. Rodder S, Scherer A, Korner M, Marti HP. A subset of metzincins and
T cell migration to vascularized transplants. J Clin Invest. related genes constitutes a marker of human solid organ fibrosis.
2013;123(6):2663‐2671. Virchows Arch. 2011;458(4):487‐496.
36. Zhuang Q, Liu Q, Divito SJ, et al. Graft-infiltrating host dendritic 46. Rodder S, Scherer A, Korner M, et al. Meta-analyses qualify metz-
cells play a key role in organ transplant rejection. Nat Commun. incins and related genes as acute rejection markers in renal transplant
2016;24(7):12623. patients. Am J Transplant. 2010;10(2):286‐297.
37. Salazar IDR, Lopez MM, Chang J, Halloran PF. Reassessing the sig- 47. Bonaccorsi-Riani E, Pennycuick A, Londono MC, et al. Molecular char-
nificance of v-lesions in kidney transplant biopsies. J Am Soc Nephrol. acterization of acute cellular rejection occurring during intentional im-
2015;26(12):3190‐3198. munosuppression withdrawal in liver transplantation. Am J Transplant.
38. Einecke G, Reeve J, Mengel M, et al. Expression of B cell and immuno- 2016;16(2):484‐496.
globulin transcripts is a feature of inflammation in late allografts. Am J 48. Saint-Mezard P, Berthier CC, Zhang H, et al. Analysis of independent
Transplant. 2008;8(7):1434‐1443. microarray datasets of renal biopsies identifies a robust transcript sig-
39. Mengel M, Reeve J, Bunnag S, et al. Molecular correlates of scarring nature of acute allograft rejection. Transpl Int. 2009;22(3):293‐302.
in kidney transplants: the emergence of mast cell transcripts. Am J 49. Wang E, Worschech A, Marincola FM. The immunologic constant of
Transplant. 2009;9(1):169‐178. rejection. Trends Immunol. 2008;29(6):256‐262.
40. Modena BD, Kurian SM, Gaber LW, et al. Gene expression in biop- 50. Spivey TL, Uccellini L, Ascierto ML, et al. Gene expression profiling
sies of acute rejection and interstitial fibrosis/tubular atrophy reveals in acute allograft rejection: challenging the immunologic constant of
highly shared mechanisms that correlate with worse long-term out- rejection hypothesis. J Transl Med. 2011;9:174.
comes. Am J Transplant. 2016;16(7):1982‐1998.
41. Hruba P, Brabcova I, Gueler F, et al. Molecular diagnostics identifies
risks for graft dysfunction despite borderline histologic changes.
How to cite this article: Halloran PF, Venner JM, Madill-
Kidney Int. 2015;88(4):785‐795.
42. Broin O, Hayde N, Bao Y, et al. A pathogenesis-based transcript sig-
Thomsen KS, et al. Review: The transcripts associated with
nature in donor-specific antibody-positive kidney transplant patients organ allograft rejection. Am J Transplant. 2018;18:785–795.
with normal biopsies. Genom Data. 2014;2:357‐360. https://doi.org/10.1111/ajt.14600
43. Khatri P, Roedder S, Kimura N, et al. A common rejection module
(CRM) for acute rejection across multiple organs identifies novel