The Trifecta Study: Comparing Plasma Levels of Donor-derived Cell-Free DNA with the Molecular Phenotype of Kidney Transplant Biopsies

Abstract

Background 

The relationship between the donor-derived cell-free DNA fraction (dd-cfDNA[%]) in plasma in kidney transplant recipients at time of indication biopsy and gene expression in the biopsied allograft has not been defined.

Methods 

In the prospective, multicenter Trifecta study, we collected tissue from 300 biopsies from 289 kidney transplant recipients to compare genome-wide gene expression in biopsies with dd-cfDNA(%) in corresponding plasma samples drawn just before biopsy. Rejection was assessed with the microarray-based Molecular Microscope Diagnostic System using automatically assigned rejection archetypes and molecular report sign-outs, and histology assessments that followed Banff guidelines.

Results 

The median time of biopsy post-transplantation was 455 days (5 days to 32 years), with a case mix similar to that of previous studies: 180 (60%) no rejection, 89 (30%) antibody-mediated rejection (ABMR), and 31 (10%) T cell–mediated rejection (TCMR) and mixed. In genome-wide mRNA measurements, all 20 top probe sets correlating with dd-cfDNA(%) were previously annotated for association with ABMR and all types of rejection, either natural killer (NK) cell–expressed (e.g., GNLY, CCL4, TRDC, and S1PR5) or IFN-γ–inducible (e.g., PLA1A, IDO1, CXCL11, and WARS). Among gene set and classifier scores, dd-cfDNA(%) correlated very strongly with ABMR and all types of rejection, reasonably strongly with active TCMR, and weakly with inactive TCMR, kidney injury, and atrophy fibrosis. Active ABMR, mixed, and active TCMR had the highest dd-cfDNA(%), whereas dd-cfDNA(%) was lower in late-stage ABMR and less-active TCMR. By multivariate random forests and logistic regression, molecular rejection variables predicted dd-cfDNA(%) better than histologic variables.

Conclusions 

The dd-cfDNA(%) at time of indication biopsy strongly correlates with active molecular rejection and has the potential to reduce unnecessary biopsies.

Clinical Trial registration number: 

NCT04239703

Measurement of plasma donor-derived cell-free DNA (dd-cfDNA), expressed as percent of the total dd-cfDNA (dd-cfDNA[%]), was developed to screen for the presence of rejection in organ transplants,^1234–5 including in the lung,^67–8 kidney,^{2,4,5,910111213–14} and heart.^15,16 The relationship of dd-cfDNA(%) to histologic TCMR has been less consistent than for antibody-mediated rejection (ABMR).¹⁷ For example, in heart transplants the mean dd-cfDNA(%) levels measured by shotgun sequencing were five-fold higher in histologic ABMR than TCMR,¹⁵ and early post-transplant injury also elevated the dd-cfDNA(%). The dd-cfDNA(%) increased slightly in apparently stable kidney transplant patients over years post-transplant,¹⁸ possibly related to atrophy fibrosis.¹⁹

The relationship between dd-cfDNA(%) and the molecular phenotype in kidney transplants can be studied using automatic genome-wide microarray measurements in the Molecular Microscope Diagnostic System (MMDx). Microarrays measure gene expression with 99% reproducibility²⁰ and interprets disease states by machine learning–derived classifiers and archetype scores. MMDx (molecular) diagnoses are assigned in one of two ways: automatic archetype assignments, which avoid all subjectivity, or by an expert report sign-out^21,22 using rejection archetypes plus classifiers and gene sets. Molecular diagnoses shows strong agreement with histology, albeit with some discrepancies,^21,22 in part due to the intrinsic variability in histology,²³ for example, in TCMR.^24,25 Some discrepancies reflect diagnostic conventions: for example, BK virus nephropathy often has molecular TCMR-like changes due to cognate T cell responses against virus or alloantigens,²⁶ but the histologic convention is to not diagnose TCMR in BK virus nephropathy, despite typical lesions.²⁶ All biopsy assessments—molecular or histologic—can be affected by sampling error, and have potential for error around boundaries of diagnostic classes.

We launched the Trifecta Study, an investigator-initiated, international prospective trial, to define the relationship between dd-cfDNA(%) at the time of indication biopsy and the genome-wide molecular findings in the biopsy, and with the diagnostic classification of the biopsy by molecular and histologic assessments. The present results are for the first 300 consecutive biopsies (94% for indications) processed by for molecular diagnoses with blood drawn for dd-cfDNA(%) just before the biopsy.

Methods

Study Population

We analyzed 300 indication kidney transplant biopsies collected prospectively from 289 consenting patients. Eight biopsies with insufficient material for microarray processing and a further eight with unsuccessful dd-cfDNA(%) analyses or unavailable blood samples were excluded from 316 total biopsies to produce the n=300 cohort (Figure 1). Reaching 300 results triggered the present analysis, as per protocol.

Figure 1.: Study design flowchart. Workflow for collection and processing of the 300 biopsies for molecular biopsy assessment and concurrent blood samples processed for percent donor-derived cell-free DNA (%dd-cfDNA) in the Trifecta Study.

A list of centers is shown in Table 1. Demographics are shown in Table 2. Biopsies were contributed by established centers under local Institutional Review Board–approved protocols (ClinicalTrials.gov number NCT04239703).

Table 1. - Trifecta Study coauthors (n=300 biopsies): Institutions and investigators that have contributed biopsies and cfDNA data

Investigators	Institution	Location
Beata Naumnik	Medical University in Bialystok	Białystok, Poland
Jonathan Bromberg	University of Maryland School of Medicine	Maryland, Baltimore
Matt Weir
Nadiesa Costa
Milagros Samaniego-Picota	Henry Ford Transplant Institute	Detroit, Michigan
Iman Francis
Anita Patel
Alicja Dębska-Ślizień	Medical University of Gdańsk	Gdańsk, Poland
Joanna Konopa
Andrzej Chamienia
Grzegorz Piecha	Medical University of Silesia	Katowice, Poland
Željka Veceric-Haler	University of Ljubljana	Ljubljana, Slovenia
Miha Arnol
Nika Kojc
Maciej Glyda	Wojewodzki Hospital	Poznan, Poland
Katarzyna Smykal-Jankowiak
Ondrej Viklicky	Institute for Clinical and Experimental Medicine	Prague, Czech Republic
Petra Hruba
Silvie Rajnochová Bloudíčkova
Janka Slatinská
Marius Miglinas	Centre of Nephrology, Vilnius University Hospital Santaros Klinikos	Vilnius, Lithuania
Marek Myślak	Pomeranian Medical University	Szczecin, Poland
Joanna Mazurkiewicz
Marta Gryczman
Leszek Domański	University Hospital n.2, Szczecin
Rajendra Baliga	Tampa General Hospital	Tampa Bay, Florida
Agnieszka Perkowska-Ptasińska	Warsaw Medical University	Warsaw, Poland
Dominika Dęborska-Materkowska
Michał Ciszek
Magdalena Durlik
Ryszard Grenda	The Children’s Memorial Health Institute
Mirosław Banasik	Medical University of Wrocław	Wrocław, Poland
Mladen Knotek	University Hospital Merkur	Zagreb, Croatia
Ksenija Vucur
Zeljka Jurekovic
Thomas Müller	University Hospital Zurich	Zurich, Switzerland
Thomas Schachtner
Andrew Malone	Washington University at St. Louis	St. Louis, MO, USA
Tarek Alhamad

Table 2. - Demographics and clinical features of the Trifecta Study (n=300) biopsy cohort

Biopsy Characteristics
Days to biopsy post-transplant
Mean	1403
Median (range)	455 (5–11504)
Days to most recent follow-up after biopsy
Mean	177
Median (range)	169 (1–526)
Indication for biopsy (% of total)
For cause	278 (94%)
Surveillance	18 (6%)
Missing	4 (1%)
Patient demographics (n=300)
Mean patient age (range)	51 (19–80)
Age > 65 years (count)	29
Mean donor age (range)	47 (6–81)
Patient sex
Male (% of known)	187 (63%)
Female (% of known)	111 (37%)
Not available	2 (1%)
Donor sex
Male (% of known)	150 (52%)
Female (% of known)	137 (48%)
Not available	13 (4%)
Patient ethnicity
Black	11
Other	288
Not available a	1
Donor type (% deceased donor transplants)	224 (90%)
Status at last follow-up
Functioning graft	247 (90%)
Graft failure/return to dialysis	23 (8%)
Patient death with functioning graft	3 (1%)
Primary disease
Diabetic nephropathy	34
Hypertension/ large vessel disease	9
Glomerulonephritis/ vasculitis	107
Interstitial nephritis/ pyelonephritis	2
Polycystic kidney disease	0
Others	99
Unknown etiology	38

^aSome centers preferred not to identify ethnicity.

Sample and Data Collection

A portion of one core (mean length 3 mm)²⁷ was immediately stabilized in RNALater and shipped to the Alberta Transplant Applied Genomics Centre (http://atagc.med.ualberta.ca) at ambient temperature for RNA extraction and processing as previously described.²⁷ Gene expression was measured using Affymetrix PrimeView microarrays (n=49,495 probe sets) unless the biopsy was inadequate for analysis (e.g., too small or degraded: eight out of 308 in this dataset, and 4% in previous studies²⁷). Of the 300 biopsies, 285 had available histologic diagnoses (nine missing, six with inadequate material for assessment). All molecular analyses and diagnoses were made without knowledge of the histology, donor-specific antibody (DSA), clinical data, or dd-cfDNA(%) results. Molecular reports were sent to the participating centers, usually within 2 working days of receiving the biopsy.

All blood samples for dd-cfDNA(%) testing were collected before the biopsies were performed to avoid any effect of the biopsy procedure on the dd-cfDNA(%) measurement.²⁸ Blood samples were drawn in two 10 ml quantities in DNA Streck tubes using 20–21 gauge needles and shipped immediately per established protocols to Natera Inc. for analysis.

Histologic and clinical data were collected at each center and submitted to the study as available. All centers followed Banff 2019 guidelines for interpreting histology. The study was not permitted to request additional local tests that were not standard of care, and the centers did not agree to provide biopsies for central histology reading because it is not standard of care.

The Trifecta Study also includes centralized DSA measurements by One Lambda Inc., which are in progress and not reported here. A preliminary DSA analysis has been reported in abstract form²⁹ and the full analysis will be described in a subsequent paper.

The MMDx

MMDx is a central biopsy diagnostic system that measures genome-wide mRNA expression and interprets the results using many previously derived and locked machine-learning–derived algorithms to assign molecular diagnoses, comparing new biopsies with results in a large reference set (n=1208). Automated output from ensembles of machine-learning algorithms (classifiers), both supervised and unsupervised,^21,22 were used to make observer-independent assessments. Outputs include assessments of parenchymal injury and atrophy fibrosis. All molecular measurements were also assessed by an expert reader, who signed out the molecular MMDx report.

Principal Component Analysis and Archetypal Analysis

Principal component analysis (PCA) is a dimensionality reduction technique used for data visualization and interpretation. Archetypal analysis (AA) is a method related to clustering that represents each sample as a composite of a small number (n) of archetypes (clusters)—theoretical constructs representing diverse “types” of samples, each type with its own characteristic set of features. AA automatically assigns scores to each biopsy, describing its relationship to each archetype group. We named these groups on the basis of their associations with molecular and histologic findings: No rejection, TCMR1 (intense TCMR, sometimes mixed with ABMR³⁰), TCMR2 (less active TCMR), early-stage ABMR (EABMR), fully developed ABMR (FABMR), and late-stage ABMR.

In this paper, PCA and AA are both constructed using the same 1208 (samples) × 7 (features) data matrix.³⁰ Each of the seven input features is a score from a classifier trained and tested in the 1208 reference set. Two of the seven inputs are classifiers predicting histologic rejection—one for ABMR and one for TCMR. The other five are for predicting rejection-related histologic lesions: i>1, t>1, ptc>0, g>0, and cg>0. We used the PCA function in the R FactoMineR package³¹ and the stepArchetypes function in the archetype package.

Lists of Genes

Significant gene lists were determined by Spearman correlation coefficients (SCC) between the expression levels and the other variable of interest. The resulting transcript lists were ordered by correlation coefficient.

The dd-cfDNA(%) Assay

The dd-cfDNA(%) assays (Prospera) were performed on blood samples drawn at time of biopsy. The results were not reported to the center until many months postbiopsy, and never influenced the decision to biopsy or patient management. The Prospera assay uses massively multiplexed targeted PCR to identify 13,392 single nucleotide polymorphisms.⁹ For analytical purposes in this study, some analyses used the cutoff of dd-cfDNA(%) ≥1%. All dd-cfDNA(%) read as “<0.08%” (n=23) are analyzed as 0.08%.

Results

Case Mix in 300 Biopsies

Molecular diagnoses, rejection archetype groups, plus local standard-of-care histology, and DSA data are given in Table 3. Case mix was similar to the indication biopsies in the reference set.²² Archetypal assignments were No rejection 60%; TCMR 10% (TCMR1 3%; TCMR2 7%); and ABMR 30% (EABMR 10%, FABMR 13%, late-stage ABMR 6%). Molecular sign-outs were No rejection 58%; TCMR 7%; possible TCMR 2%; mixed 6%; ABMR 22%; and possible ABMR 4%. Histology assigned No rejection 43%; TCMR 10%; possible TCMR (borderline) 10%, ABMR 15%, possible ABMR 12%, and mixed 5%, with 5% inadequate for reporting.

Table 3. - Histologic diagnoses and DSA status in the Kidney 300 cohort (n, % of total)

	Molecular Diagnoses		Histology Diagnoses
	n (% of 300)	% DSA Positive of Those Tested	n (% of 300)	% DSA Positive of Those Tested
No rejection	175 (58) a	6	128 (43) a	9
Possible TCMR (pTCMR)	6 (2)	0	31 (10)	10
TCMR	21 (7) a	29	31 (10) a	19
Possible ABMR (pABMR)	12 (4)	50	35 (12)	26
ABMR	67 (22) a	33	45 (15) a	44
Mixed (ABMR+TCMR)	19 (6) a	50	15 (5) a	40
All ABMR (including Mixed)	86 (29)	36	60 (20)	37
Inadequate	—	—	6 (2)	—
Missing	—	—	9 (3)	—

Automated rejection archetypes
	n (% of 300)		% DSA positive of those tested
No rejection	180 (60) a		20
TCMR1 (many mixed)	9 (3) a		50
TCMR2	22 (7)		29
EABMR	31 (10) a		44
FABMR	40 (13) a		58
LABMR	18 (6)		38

LABMR, late-stage ABMR.

^aGroups with specific references in text.

The dd-cfDNA(%) levels rose slightly with time of biopsy post-transplant (linear regression P=0.06). The “supersmoother” function shows these levels are also somewhat elevated in biopsies in the first weeks post-transplant (Figure 2). Very low levels of dd-cfDNA(%) were uncommon in early biopsies, presumably due to AKI resulting from recent donation-implantation.³² Most ABMR biopsies (blue) were late, whereas most early biopsies were No rejection (gray).

Figure 2.: Relationships between %dd-cfDNA, time of biopsy post-transplant, and molecular archetype groups in N=300 samples. Dots represent biopsies and corresponding paired blood sample %dd-cfDNA results, colored by archetype cluster assignments. Relationships between %dd-cfDNA and time post-transplant are represented by linear regression (thick dashed black line), restricted cubic spline (with three knots - blue line) and a supersmoother (red line). Overall, %dd-cfDNA increases slightly over time post-transplant (linear regression P=0.06). %dd-cfDNA ≥ 1 is used as the cutoff throughout this paper for positivity. NR – no rejection.

Top Transcripts Associated with dd-cfDNA(%)

Table 4 shows the top 20 transcripts correlated with dd-cfDNA(%); all were positively correlated. (Negative correlations were much weaker.) All of the top 20 (of 49,495 total probe sets) had previously been annotated as being highly associated with all types of rejection and with ABMR, particularly representing NK cell genes (e.g., GNLY, CCL4, TRDC, and S1PR5) and IFNG-inducible genes (e.g., PLA1A, IDO1, CXCL11, WARS). The top transcript was granulysin (GNLY), with a SCC of 0.56 (P=4E-26). Analysis of the top 20 unique genes correlated with dd-cfDNA(%) was similar (Supplemental Table 1): 19 had previously been annotated as expressed by NK cells (11) or IFNG-inducible (eight), and one had been annotated as inducible by injury.

Table 4. - Top probe sets positively correlated with dd-cfDNA(%) in the Trifecta Study (n=300) ordered by correlation coefficient

Affy ID	Gene Symbol	Gene Name	Transcript Set Annotation	SCC a
11751857_a_at	GNLY b	granulysin	ABMR-RAT, Rej-RAT	0.56
11763715_a_at	GNLY b	granulysin	ABMR-RAT,Rej-RAT	0.55
11746954_s_at	CCL4 b	chemokine (C-C motif) ligand 4	ABMR-RAT,Rej-RAT	0.52
11761790_x_at	TRDC b	T cell receptor delta constant	ABMR-RAT,Rej-RAT	0.51
11718982_s_at	CCL4 b	chemokine (C-C motif) ligand 4	ABMR-RAT,Rej-RAT	0.51
11718983_x_at	CCL4 b	chemokine (C-C motif) ligand 4	ABMR-RAT,Rej-RAT	0.51
11743168_at	IDO1 c	indoleamine 2,3-dioxygenase 1	ABMR-RAT,GRIT3,Rej-RAT	0.51
11749245_a_at	CXCL11 c	chemokine (C-X-C motif) ligand 11	ABMR-RAT,GRIT3,Rej-RAT	0.50
11753484_x_at	KLRD1 b	killer cell lectin-like receptor subfamily D, member 1	ABMR-RAT,Rej-RAT	0.50
11727116_a_at	PLA1A c	phospholipase A1 member A	ABMR-RAT, Rej-RAT	0.50
11729649_at	PRF1 b	perforin 1 (pore forming protein)	ABMR-RAT,QCAT,Rej-RAT	0.50
11740452_x_at	KLRD1 b	killer cell lectin-like receptor subfamily D, member 1	ABMR-RAT,Rej-RAT	0.50
11724900_a_at	GZMB b	granzyme B	ABMR-RAT,QCAT,Rej-RAT	0.49
11726287_a_at	WARS c	tryptophanyl-tRNA synthetase	ABMR-RAT,GRIT3,Rej-RAT	0.49
11733353_at	CRTAM b	cytotoxic and regulatory T-cell molecule	ABMR-RAT,RAT,Rej-RAT	0.49
11756632_a_at	GNLY b	granulysin	ABMR-RAT,QCAT,Rej-RAT	0.49
11744660_s_at	CCL4L1 b	chemokine (C-C motif) ligand 4-like 1	ABMR-RAT, Rej-RAT	0.49
11732466_a_at	CXCL11 c	chemokine (C-X-C motif) ligand 11	ABMR-RAT,GRIT3,Rej-RAT	0.49
11752664_a_at	S1PR5 b	sphingosine-1-phosphate receptor 5	ABMR-RAT,RAT	0.49
11763756_x_at	GNLY b	granulysin	ABMR-RAT,QCAT,RAT	0.49

Transcript sets are described on the home page at https://www.ualberta.ca/medicine/institutes-centres-groups/atagc/research/gene-lists.html. ABMR-RAT, antibody-mediated rejection-associated transcripts; GRIT, gamma-interferon-inducible transcripts, QCAT, Cytotoxic T cell-associated transcripts; Rej-RAT, rejection-associated transcripts.

^aAll P values were <0.0001.

^bNK cell expressed.

^cIFNG-inducible.

Supplemental Figure 1 shows the correlation between dd-cfDNA(%) and the top dd-cfDNA(%)–associated gene, GNLY, with each biopsy colored by its archetypal cluster assignment, showing the strong association of EABMR and FABMR (cyan and blue symbols, respectively) with high GNLY expression and dd-cfDNA(%). Many archetypal TCMR1 biopsies (red) and some TCMR2 (green) also had high GNLY expression and dd-cfDNA(%).

Associations of Classifier and Transcript Set Scores with dd-cfDNA(%)

Table 5 analyzes the scores for previously annotated transcript sets and classifiers for their SCC with dd-cfDNA(%) in the 300 biopsies. The highest correlation was with the classifier trained on peritubular capillary inflammation, ptc_Prob (SCC 0.54). Other high correlations were related to scores for all-rejection or ABMR: the ABMR_Prob, g_Prob, ptc_Prob, and Rej_Prob classifiers, and the ABMR-RAT transcript set. The correlations with scores for TCMR gene sets and classifiers were also high, although weaker than the associations with ABMR and all-rejection. Injury and atrophy-fibrosis gene set and classifier scores were weakly associated with dd-cfDNA(%), and the normal kidney transcript set score correlated negatively, reflecting injury-induced parenchymal dedifferentiation.

Table 5. - SCC of molecular rejection and injury transcript set and classifier scores with dd-cfDNA(%) in n=300 biopsies

Variable			SCC	P Value
ABMR related	ABMR probability classifier	ABMRProb a	0.52 a	<0.0001 a
	cg>0-probability classifier	cg>0Prob	0.35	<0.0001
	g>0-probability classifier	g>0Prob a	0.50 a	<0.0001 a
	ptc>0 probability classifier	ptc>0Prob a	0.54 a	<0.0001 a
	DSA-probability classifier	DSAProb	0.39	<0.0001
	ABMR transcripts	ABMR-RAT a	0.51 a	<0.0001 a
All rejection	Rejection-probability classifier	RejProb a	0.52 a	<0.0001 a
	IFNG-inducible transcripts	GRIT3	0.44	<0.0001
TCMR related	Effector T cell transcripts	QCAT	0.43	<0.0001
	TCMR-probability classifier	TCMRProb	0.22	<0.0001
	i>1-probability classifier	i>1Prob	0.34	<0.0001
	t>1-probability classifier	t>1Prob	0.26	<0.0001
Macrophage related	Macrophage transcripts	QCMAT	0.37	<0.0001
	Alternative macrophage activation transcripts	AMAT1	0.39	<0.0001
Recent injury	Recent injury-induced	IRRAT30	0.18	0.002
	Injury-induced day 3	IRITD3	0.18	0.001
Parenchymal	Normal kidney transcripts	KT1	−0.19	0.0001
Atrophy fibrosis	ci>1-probability classifier	ci>1Prob	0.16	0.005
	ct>1-probability classifier	ct>1Prob	0.13	0.02

Transcript sets are described on the home page at https://www.ualberta.ca/medicine/institutes-centres-groups/atagc/research/gene-lists.html.

^aClassifiers and gene sets with correlation >0.50.

Figure 3 shows the relationship between dd-cfDNA(%) for each biopsy and the scores for the ABMR classifier (Figure 3A) or the TCMR classifier (Figure 3B). The regression lines showing best fit within the data demonstrated significant relationships with the ABMR_Prob and TCMR_Prob classifiers. The SCC of dd-cfDNA(%) with the ABMR classifier (0.52, P=6E-22) was higher than correlation with the TCMR classifier (0.22, P=9E-5), although this will be influenced by the greater abundance of ABMR in the case mix.

Figure 3.: Relationships between %dd-cfDNA, molecular archetype groups, and the ABMR _Prob and TCMR _Prob classifier scores in N =300 samples. Dots represent biopsies and corresponding paired blood sample %dd-cfDNA results, colored by archetype cluster assignments. Regression lines (dashed) show the relationship between the (A) ABMR_Prob and (B) TCMR_Prob classifier scores and %dd-cfDNA. Spearman correlations with dd-cfDNA were stronger for ABMR_Prob (0.52, P=6E-22) than TCMR_Prob (0.22, P=9E-5). NR, no rejection.

PCA of Rejection

We used PCA to visualize the relationship between dd-cfDNA(%) and the seven rejection classifiers used to assign archetype scores²¹ (Figure 4). When we projected dd-cfDNA(%) scores into the PCA as a supplementary variable (not used as input), the dd-cfDNA(%) vector correlated positively with PC1 (all-rejection) and PC2 (ABMR, Figure 4A). In PC3, dd-cfDNA(%) performed like the active ABMR-related classifiers, reflecting association with early-stage or fully-developed ABMR rather than late-stage ABMR (Figure 4B). The dd-cfDNA(%) vector strongly resembled the peritubular capillaritis (ptc >1_Prob) molecular classifier vector in all three dimensions.

Figure 4.: Rejection classifier PCA factor maps in N=300 samples, showing %dd-cfDNA as a supplementary variable. (A) PC2 versus PC1. (B) PC2 versus PC3. Input variables (classifier scores) are shown as solid black lines, and %dd-cfDNA in dashed grey as it is not included as input in the PCA. %dd-cfDNA behaves like a molecular rejection measurement, more closely aligned to ABMR than to TCMR.

dd-cfDNA(%) in Each Diagnostic Group

Among the archetype groups illustrated in Figure 5 the dd-cfDNA(%) was highest in FABMR, EABMR, TCMR1, and some TCMR2 biopsies. A small number of no rejection biopsies also had high values. Note that late-stage ABMR (which is often inactive³⁰) often had low dd-cfDNA(%).

Figure 5.: Beeswarm/boxplots showing %dd-cfDNA measurements versus molecular rejection archetype groups in N =300 samples. Dots represent biopsies and corresponding paired blood sample %dd-cfDNA results.

Table 6 shows the mean and median dd-cfDNA(%) within each of the diagnostic categories: molecular sign-outs, automated rejection archetypes, and histology sign-outs. Because of the skewed distribution of the dd-cfDNA(%) data, medians are a more reliable measure of central tendency than means, thus we focus on medians.

Table 6. - Mean and median dd-cfDNA(%) levels in biopsies grouped by rejection diagnoses in n=300 biopsies

Variables			dd-cfDNA(%)		n
			Mean	Median
Molecular sign-outs	No rejection		0.62	0.33	175
	TCMR related	TCMR (“pure” TCMR) a	2.27	1.61	21
		pTCMR	0.55	0.27	6
	Mixeda		2.84	1.56	19
	ABMR related	ABMR a	3.12	2.11	67
		pABMR	0.85	0.57	12
Automated rejection archetypes	No rejectiona		0.70	0.32	180
	TCMR related	TCMR1 (including mixed) a	3.49	3.71	22
		TCMR2 a	2.08	1.26	9
	ABMR related	EABMR a	2.29	1.80	31
		FABMR a	3.54	2.12	40
		LABMR	0.99	0.54	18
Histology diagnostic sign-outs by the centers	No rejection	“AKI” (<6 weeks post-transplant) a	0.90	0.68	14
		>6 weeks post-transplant a	0.24	0.16	25
		Other	0.72	0.36	89
	TCMR related	TCMR (“pure” TCMR) a	1.53	0.88	31
		pTCMR (borderline)	0.78	0.26	31
	ABMR related	ABMR a	2.16	1.57	45
		pABMR	2.78	1.14	35
	Mixed (ABMR+TCMR) a		3.83	3.18	15
	Inadequate		2.54	1.23	6
	Missing		1.20	0.26	9

LABMR, late-stage ABMR.

^aThe principal diagnostic categories.

Among molecular sign-out categories, median dd-cfDNA(%) was highest in ABMR, and mixed rejection, moderately high in TCMR, and lowest in No rejection.

Among rejection archetypes, median dd-cfDNA(%) was highest in TCMR1 (which includes some mixed) and lowest in No rejection.

In histology sign-outs, median dd-cfDNA(%) was highest in ABMR and mixed rejection biopsies; TCMR had moderate elevations. Biopsies with no histologic abnormalities, >6 weeks post-transplant (i.e., after recovery of AKI from transplantation, n=25) had the lowest median dd-cfDNA(%) (0.16), whereas biopsies with no histologic abnormalities <6 weeks post-transplant (n=14) had mildly elevated values (0.68, P=0.0003).

Note that “pure” TCMR identified by molecular sign-out diagnoses and by histology diagnoses had elevated mean and median dd-cfDNA(%) compared with No rejection biopsies.

Variable Importance in Predicting dd-cfDNA(%) ≥1

Figure 6 shows the relative variable importance for predicting dd-cfDNA(%) ≥1% using random forest analysis. In Figure 6A, the molecular classifiers and gene sets predicted dd-cfDNA(%) ≥1% with error rate of 0.20, with the top predictors being the molecular ptc-lesion classifier (ptc_Prob) and the all-rejection (Rej_Prob) classifier. In general, ABMR-related (blue) and all rejection-related (purple) molecular features were the most important predictors of dd-cfDNA(%).

Figure 6.: Random forests demonstrating predictive ability of various molecular and histologic features of rejection for %dd-cfDNA ≥ 1.0 in N =300. (A) Relative variable importance using molecular predictors. (B) Relative variable importance using histologic predictors. In both figures, ABMR and overall rejection related variables are the most important predictors of high %dd-cfDNA.

In Figure 6B, the same analyses were repeated for histologic variables plus local DSA (error rate=0.30). The top variable was the histology ptc-lesion score. The overall prediction of dd-cfDNA(%) was more accurate with molecular variables than with histology variables and DSA (error rates: 0.20 versus 0.30, area under the receiver operating characteristic curves for the out-of-bag (bootstrapped) predicted values: 0.80 versus 0.71, P=0.003). DSA status was relatively unimportant compared with biopsy features, and excluding the local DSA status as a variable did not affect the predictions (data not shown).

Thus, the molecules associated with ptc lesions (as estimated by the ptc_Prob classifier), and to a lesser extent the ptc-lesions themselves, are the biopsy features in the biopsy with the strongest association with dd-cfDNA production.

Logistic regression models predicting dd-cfDNA(%) ≥1.0 were built using the top five variables from each random forest analysis. Models used histology variables alone, molecular variables alone, or all 10 combined. The combined model added significantly to the predictive value of histology alone (P=2×10⁻⁸, likelihood ratio test) but not to molecules alone (P=0.99). The same comparisons were run using linear regression to predict log₁₀(dd-cfDNA[%]) as a continuous variable. Again, molecules added to histology alone (P=2×10⁻¹¹), but histology did not add to molecules alone (P=0.95). Random forest models are not easily comparable with methods such as likelihood ratio tests, but the area under the receiver operating characteristic curves from their predicted values can be compared.

Discussion

This study prospectively recruited consenting patients undergoing kidney transplant indication biopsies (94% recorded as being for indications) to examine the relationship between dd-cfDNA(%) results taken just before the biopsy and gene expression in the biopsy. The 300 biopsies represent a cross-section of the international kidney transplant population undergoing indication biopsies from 5 days to 32 years post-transplant (median 453 days). The case mix was similar to previous indication biopsy studies, with approximately 60% of biopsies assigned molecular No rejection and 30% molecular ABMR by molecular sign-outs or automatic archetypes. Many of the elevated dd-cfDNA(%) values after 1 year had ABMR. All top 20 probe sets correlated with dd-cfDNA(%) were associated with all-rejection and/or with ABMR and expressed in NK cells or IFNG-inducible. The strongest correlation of multigene measurement scores (transcript sets or classifiers) with dd-cfDNA(%) were similar—strongest with ABMR and all-rejection scores, moderately with TCMR scores, and weakly with recent parenchymal injury, dedifferentiation, and atrophy-fibrosis scores. Using PCA, dd-cfDNA(%) performed as a variable associated with rejection, but specifically with active ABMR (EABMR and FABMR) rather than late-stage ABMR. By diagnostic categories, dd-cfDNA(%) was highest in both molecular and histologic ABMR, increased in TCMR biopsies, and mildly increased in AKI and atrophy fibrosis compared with histologically normal biopsies after 6 weeks post-transplant. The associations of dd-cfDNA(%) were strong with both molecular and histologic rejection, and were somewhat stronger with the molecular measurements, similar to the recent findings in an independent cohort.³³ We conclude that dd-cfDNA(%) is strongly associated with active rejection, particularly ABMR and mixed but also the most active TCMR.

The top genes correlating with circulating dd-cfDNA(%) at the time of indication biopsies are predominantly those associated with active ABMR—genes expressed by NK cells or inducible by IFNG. These top transcripts are commonly expressed in active and fully developed ABMR in kidney and heart transplants.^30,34 However, active ABMR is much more common than active TCMR in this population, and active pure TCMR also releases dd-cfDNA.

Our hypothesis is that the degree of dd-cfDNA(%) released by TCMR reflects the activation state of the effector T cells in those TCMR biopsies. TCMR varies in molecular activity, as illustrated by the TCMR1 and TCMR2 archetypes groups. TCMR1 often has ABMR features, but always has more intense TCMR activity (e.g., IFNG expression) than TCMR2.³⁰ Some TCMR-like states recognized by MMDx (particularly TCMR2) or by histology are relatively inactive in molecular findings. Although associated ABMR contributes to the high dd-cfDNA(%) in some TCMR1 biopsies (e.g., TCMR1), it is likely that active pure TCMR also releases dd-cfDNA without ABMR activity, and TCMR biopsies associated with high dd-cfDNA(%) cannot be explained by associated ABMR activity. Our hypothesis is that some TCMR phenotypes release lower amounts of dd-cfDNA because of a lower activation state of the cognate T cells in these patients, due either to effects of immunosuppressive drugs or to adaptive changes in the T cells, such as exhaustion. Adaptive changes presumably explain why TCMR episodes become uncommon in transplant populations after 5–10 years.³⁵ The potential relationship of dd-cfDNA(%) to cognate T cell activity within TCMR biopsies (as opposed to undetected ABMR activity) will be a subject of ongoing investigation in the Trifecta study.

The dd-cfDNA(%) test is of interest because of its potential to avoid unnecessary indication biopsies when dd-cfDNA(%) is low in patients with biopsy indications, although whether it actually does avoid unnecessary biopsies has not yet been established and was not examined in this study. (The use of dd-cfDNA[%] as a screening test in patients without biopsy indications is a separate issue and not addressed in this indication biopsy study.) The highest dd-cfDNA(%) values were typically associated with molecularly active rejection, and very low values were associated with pristine biopsies that did not have molecular or histologic rejection or other abnormalities. (Pristine biopsies were those that had recovered from the injury induced by the transplantation process, >6 weeks post-transplant, and without any signs of current rejection.) Although the dd-cfDNA test (Prospera) is typically only validated in patients meeting a set of criteria (e.g., >14 days post-transplant, no active cancer), a small number of biopsies in this study (approximately 10%) were taken from patients that would be excluded by these criteria. We designed the Trifecta study to find the associations in all indication biopsies with no selections, which permits us to draw conclusions from the largest possible dataset and can potentially inform future decisions for clinical use.

The explanation for the release of dd-cfDNA(%) by some kidneys with no molecular or histologic rejection could reflect parenchymal injury (acute injury or atrophy fibrosis), although low-level subthreshold ABMR should be considered.³⁶ There is also a weak but significant association of dd-cfDNA(%) with AKI (e.g., in early transplants) and with atrophy fibrosis, and the source of the dd-cfDNA needs to be defined as potentially useful information about the renal parenchyma.

This study is limited in numbers of biopsies within individual rejection categories, which will be addressed in the ongoing extension of the study. One limitation of this study is that we did not include centralized reading of the biopsy histology, which would be of interest in addition to molecular readings and local histology readings.

In summary, Trifecta finds that dd-cfDNA(%) can be useful at the time of an indication biopsy—if dd-cfDNA(%) is very low, it means the probability of active rejection is low. In PCA, the orientation of dd-cfDNA(%) in blood was similar to the measurements in the biopsy itself, which is good evidence that plasma dd-cfDNA(%) is a useful addition to existing clinical information. In addition to guiding decisions at indication biopsy, and as part of routine screening, dd-cfDNA(%) could potentially be used to follow response-to-treatment, and avoid follow-up biopsies if the dd-cfDNA(%) signal normalizes. However, the potential benefits of these applications need to be proven.

Disclosures

A. Prewett reports current employment with, and an ownership interest in, Natera. K. Madill-Thomsen reports current employment with Transcriptome Sciences Inc., which is a University of Alberta Research Company; and reports receiving research funding from the Alberta Ministry of Advanced Education and Technology, the Canada Foundation for Innovation, Genome Canada, Industrial Research Assistance Program, the Mendez National Institute of Transplantation Foundation, the University of Alberta Hospital Foundation; partial support was also provided by funding from a licensing agreement with the One Lambda division of Thermo Fisher Scientific. P. Billings reports current employment with Natera Inc.; having consultancy agreements with Alveo Technologies, Baebies Inc., OmniSeq Inc., and Target BioPharma Inc.; reports having an ownership interest in 23andMe, Biological Dynamics, DNArx Inc., Fabric Genomics Inc., Iliad Bioscience Inc., LabCorp Inc., LungLifeAI Inc., MissionBio Inc., Natera Inc., OmniSeq Inc., PlumCare LLc; Precision Biosciences Inc., Roswell Biosciences, ThermoFisher Inc.; and reports being a scientific advisor or member of Alveo Technologies, Baebies Inc., Biological Dynamics, Fabric Genomics Inc., LungLifeAI Inc., MissionBio Inc., OmniSeq Inc., PlumCare LLc, and TargetBioPharma, Inc. The Trifecta study is an investigator-initiated study supported by a grant from Natera to Transcriptome Sciences Inc./Alberta Transplant Applied Genomics Centre. P. F. Halloran reports having shares in Transcriptome Sciences Inc., a University of Alberta research company with an interest in molecular diagnostics, and is a consultant to Natera; all Natera authors are employees and own equity at Natera, Inc.; reports having consultancy agreements with Astellas, and CSL Behring; reports having an ownership interest in giving lectures for Thermo Fisher Scientific; reports receiving research funding from Alberta Ministry of Advanced Education and Technology, Canada Foundation for Innovation, Genome Canada, Industrial Research Assistance Program the Mendez National Institute of Transplantation Foundation, and the University of Alberta Hospital Foundation, partial support was also provided by funding from a licensing agreement with the One Lambda division of Thermo Fisher Scientific; reports being a Canada Research Chair in Transplant Immunology; reports receiving honoraria from Astellas, Natera Inc., and One Lambda; reports being a scientific advisor or member as CEO, Transcriptome Sciences Inc., Editor-in-chief emeritus of the American Journal of Transplantation; and reports speakers bureau from Astellas, Natera Inc., and One Lambda. Z. Demko is employed by and owns stock and options to buy stock in Natera. All remaining authors have nothing to disclose.

Funding

The Microarray biopsy assessment project is supporte d in part by a licensing agreement with One Lambda/Thermo Fisher Scientific.