CLINICAL RESEARCH
www.jasn.org
Cell-Free DNA and Active Rejection in Kidney Allografts
Roy D. Bloom,* Jonathan S. Bromberg,
†
Emilio D. Poggio,
‡
Suphamai Bunnapradist,
§
Anthony J. Langone,
|
Puneet Sood,
¶
Arthur J. Matas,** Shikha Mehta,
††
Roslyn B. Mannon,
††‡‡
Asif Sharfuddin,
§§
Bernard Fischbach,
||
Mohanram Narayanan,
¶¶
Stanley C. Jordan,
§
*** David Cohen,
†††
Matthew R. Weir,
‡‡‡
David Hiller,
§§§
Preethi Prasad,
|||
Robert N. Woodward,
¶¶¶
Marica Grskovic,
¶¶¶
John J. Sninsky,
¶¶¶
James P. Yee,
|||
and Daniel C. Brennan,**** for the Circulating Donor-Derived Cell-Free DNA in
Blood for Diagnosing Active Rejection in Kidney Transplant Recipients (DART) Study Investigators
*Department of Medicine, University of Pennsylvania, Perelman School of Medicine and Penn Kidney Pancreas Transplant
Program, Philadelphia, Pennsylvania;
†
Department of Surgery and Department of Microbiology and Immunology and
‡‡‡
Division of Nephrology, Department of Medicine, University of Maryland School of Medicine, Baltimore, Maryland;
‡
Department of Nephrology and Hypertension, Cleveland Clinic, Cleveland, Ohio;
§
Department of Medicine, David Geffen
School of Medicine at the University of California Los Angeles, Los Angeles, California;
|
Department of Medicine, Vanderbilt
University Medical Center, and Medical Specialties Clinic, Veteran Affairs Hospital Renal Transplant Program, Nashville,
Tennessee;
¶
Thomas Starzl Transplant Institute, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania; **Division
of Transplantation, Department of Surgery, University of Minnesota, Minneapolis, Minnesota;
††
Division of Nephrology,
Department of Medicine, and
‡‡
Division Transplantation, University of Alabama School of Medicine, Birmingham, Alabama;
§§
Division of Nephrology and Transp lant, Department of Medicine, Indiana University School of Medicine, Indianapolis, Indiana;
||
Baylor Research Institute, Dallas, Texas;
¶¶
Division of Nephrology & Hypertension, Texas A&M Health Science Center College
of Medicine, Temple, Texas; ***Division of Nephrology, Cedars-Sinai Medical Center, Los Angeles, California;
†††
Department of
Surgery, Columbia University Medical Center, New York, New York;
§§§
Biostatistics,
|||
Clinical Research,
¶¶¶
Research and
Development, CareDx, Inc., Brisbane, California; and ****Washington University School of Medicine, St. Louis, Missouri
ABSTRACT
Histologic analysis of the allograft biopsy specimen is the standard method used to differentiate rejec tion from other
injury in kidney transplants. Donor-derived cell-free DNA (dd-cfDNA) is a noninvasive test of allograft injury that may
enable more frequent, quantitative, and safer assessment of allograft rejection and injury status. To investigate this
possibility, we prospectively collected blood specimens at scheduled intervals and at the time of clinically indicated
biopsies. In 102 kidney recipients, we measured plasma levels of dd-cfDNA and correlated the levels with allograft
rejection status ascertained by histology in 107 biopsy specimens. The dd-cfDNA level discriminated between
biopsy specimens showing any rejection (T cell–mediated rejection or antibody-mediated rejection [ABMR]) and
controls (no rejection histologically), P,0.001 (receiver operating characteristic area under the curve [AUC], 0.74;
95% confidence interval [95% CI], 0.61 to 0.86). Positive and negative predictive values for active rejection at a cutoff
of 1.0% dd-cfDNA were 61% and 84%, respectively. The AUC for discriminating ABMR from samples without ABMR
was 0.87 (95% CI, 0.75 to 0.97). Positive and negative predictive values for ABMR at a cutoff of 1.0% dd-cfDNA were
44% and 96%, respectively. Median dd-cfDNA was 2.9% (ABMR), 1.2% (T cell–mediated types $IB), 0.2% (T cell–
mediated type IA), and 0.3% in controls (P=0.05 for T cell–mediated rejection types $IB versus controls). Thus, dd-
cfDNA may be used to assess allograft rejection and injury; dd-cfDNA levels ,1% reflect the absence of active
rejection (T cell–mediated type $IB or ABMR) and levels .1% indicate a probability of active rejection.
J Am Soc Nephrol 28: 2221–2232, 2017. doi: https://doi. org/10 .1681/ASN.2016091034
Received September 26, 2016. Accepted January 6, 2017.
Published online ahead of print. Publication date available at
www.jasn.org.
Correspondence: Dr. Daniel C. Brennan, Division of Nephrology,
Washington University School of Medicine, 660 S. Euclid Ave-
nue, Campus Box 8126, St. Louis, MO. Email: dbrennan@dom.
wustl.edu
Copyright © 2017 by the American Society of Nephrology
J Am Soc Nephrol 28: 2221–2232, 2017 ISSN : 1046-6673/2807-2221
2221
Accurate and ti mely detection of allogra ft rejecti on and effec-
tive treatment are essential for long-term surv ival of renal
transplants. Althoug h histology obtained via needle biopsy
remains the standard for di agnosis of rejection, this technique
is infrequently used for surveillance because of the cost, logis-
tics, pote ntial compli cations, and patien t discomfor t and
inconvenience. Donor-der ived cell-free D NA (dd-cfDNA) de-
tected in th e blood of transplant recipients has been proposed
as a noninvasi ve marker for diagnosis o f graft rejection.
1–3
The
premise for quantitative interpretation of this b iomarker is
that rejection entails in jury, including increased cell death in
the allograft, leading to increased dd-cfDNA released into the
bloodstream.
Data from several sing le-center studies suggest that
dd-cfDNA levels in blood, measured as a fraction of the total
cell-free DNA (cfDNA), can discriminate rejection from non-
rejection in heart, lung, liver, and kidney allografts. In stabl e
heart transplant recipients, the fraction of cfDNA originating
from the graft is nearly always ,1%,
4–6
where as during re-
jection the levels of dd-cfDNA are significantly higher.
5,7
In
stable lung and liver transplant recipients, the level of dd-
cfDNA is higher than in stable heart transplant recipients,
and it further increases in moderate-to-severe rejection.
8,9
Up to now, dd-cfDNA has been least studied in renal trans-
plants; levels in stable kidney recipients are similar to those in
heart transplant recipients,
4,10
and an alyses of individual pa-
tients and a small single-center study identified hi gher levels
during biopsy-proven acute rejection.
11
In kidney transplantation, there are no existing biomar kers
that adequately measure the status of active injury to the allo-
gr aft. Serum creatinine allows estimat ion of the GFR, but it is
not specific or sensitive for allograft injury, and may not dis-
tinguish acute from c hronic loss o f function.
12,13
This report from the Circul ating Donor-Derived Cell-Free
DNA in Blood for Diagnosing Acute Rejectio n in Kidney Trans-
plant Recipients (DART) s tudy (ClinicalTrials.gov Identifier:
NCT02424227) valid ates that plasma levels of dd-cfDNA can
discriminate active rejection status. The DARTstudy is the first
multicenter study of renal allograft recipients using an analyt-
ically validated dd-cfDNA test
7
that employs targeted ampli-
fication and sequencing of single-nucleotide polymorph isms
to quan tify donor a nd recipient DNA contribution s, with-
out the need for pr ior genot y ping of donor o r recipient
DNA (AlloSure).
RESULTS
Patients, Biopsies, and Blood Samples
From April of 2015 until May of 2016, 384 renal transplant
patients were enrolled (245 w ithin 1–3monthsoftheirkidney
transplantation and 139 at the time of a clinically indicated
renal biopsy) from 14 clinical sites. Figures 1 and 2 show the
pathologists’ diag nostic findings for the 107 clinically indi-
cated biopsies that had matched plasma d d-cfDNA results.
This subset provi des the core dataset used for the analyses of
dd-cfDNA to discriminate rejection from no re jection status
(using the biopsy-based pathologists’ reports as the diagnostic
standard).
The patient characteristics of the study cohort are shown in
Table 1. The DART study population was representative of the
United St ates renal transplan t registr y population (Supple-
mental Table 1). The active re jection subgroup contained a
higher proportion of black and deceased donor organ recipi-
ents than the group without active rejection and the overall
DART popul ation. Patien ts with active rejection were also sig-
nificantly younger than patients with no rejection.
At the time of data lock, 219 patients had at least one renal
biopsy; 242 biopsies had sufficient specimens and associated
pathologists’ reported results (Figure 2) . The majority of bi-
opsies (204 of 242) we re performed for clinical suspi cion of
rejection, 34 for surveillance, and f our for follow-up of treated
rejection. Only one of 34 (3%) surveillance biopsies revealed
rej ect ion (Su ppl eme nta l Tabl e 2). Th erefo re, we di d no t
calculate the performance characteristics for dd-cfDNA to dis-
crimin ate active rejection in the scenario of n o clinical indi-
cation for biopsy.
Our primary analyses i n this study combined three sub-
classes of rejection ( T cell–mediated rejection [TCMR],
“acute/active” antibody-mediated rejection [ABMR], and
“chronic, active” ABMR) defined by the Banff working
groups
14,15
because they share some common histologic cri-
teria and the related cell injury manifestations have p otential
to involve active cell injury and death,
16
and therefore result in
increased levels of dd-cfDNA (Supplemen tal Figure 1). We use
the te rm active rejection to describe these rejection sub classes
and distinguish them from all other biopsy-based diagnoses
not pheno typically associated with active rejection (detail s in
Concise Methods).
A diagnosis of active rejection was confirmed in review of 59
pathologists’ biopsy reports: 58 cases of active rejection in 20 4
biopsies, perfo rmed for clinical suspicion, most commonly an
elevation in serum creatinine, and one case of active rejection
in 34 su rveillance biopsies. The types of active rejection are
summarized in Supplemental Table 2.
dd-cfDNA Levels in Blood Plasma
To define the area under the curve–receiver operating charac-
teristic (AUC-ROC) performance of dd-cfDNA, we included
all dd-cfDNA results that were collected at the sa me time that a
clinically indicated biopsy was performed. There were 27 bi-
opsy specimens from 27 patients with, an d 80 biopsy speci-
mens from 75 patients without, ac tive rejection . A correlation
mat rix of Banff elementary l esions and clinical features for the
107 samples are rank-ordered and color-coded by dd-cfDNA
level in Figure 1. Samples with .1% dd-cfDNA o ccurred sig-
nificantly more often (P,0.01) in the following types of re-
jection and subelements: acute/active ABMR; chronic, active
ABMR; any or moderate microvascular inflammation;
linear C4d staining in peritubular capillaries; and presence
2222 Journal of the American Society of Nephrol ogy J Am Soc Nephrol 28: 2221 –2232, 2017
CLINICAL RESEARCH
www.jasn.org
of donor-specific antibody. The dd-cfDNA thres hold of 1%
also discr i minated t y p e IB TCMR (P=0.01) and transplant
glomerulopathy (P=0.03). We computed the AUC-ROC per-
formance of serum crea tinine on this set. For estimating the
positive predictive value (PPV) and negative predictive value
(NPV) of dd-cfDNA to predict active rejection versus no a ctive
rejection, we used the prevalence of 58 active rejections in the
170 patients with 204 biopsy reports available from clinically
indicated biopsies (Figure 2).
The fraction of dd-cfDNA in blood plasma diff ered signif-
icantly between th e groups (Figure 3A). The median level of
dd-cfDNA in patien ts wi th active rejection was significantly
higher (1.6%) than in the comparator group (0. 3%) of biopsy
specimens w ithout active rejection (P,0.001). Median dd-
cfDNA levels var ied by t ype of active rejection: 2.9%
(ABMR), 1.2% ( TCMR only, ty pes IB and IIA), 0.2%
(TCMR only type IA). Because of small numbers, comparison
ofTCMRtypesincludedthecasesofmixedTCMRand
ABMR. Figure 4B shows the data for TCMR $types IB
(P=0.05 versus no active rejection) an d TCMR type IA.
The fractions of true and false positive results for dd-cfDNA
to discriminate active rejection are shown in Figure 3C. The
area under the curve (AUC) was 0.74 (95% confidence in terval
[95% CI], 0.61 to 0.86). With a cutoff of 1.0%, dd-cfDNA had
an 85% speci ficity (95% CI, 79% to 91%) and 59% sensitivity
(95% CI, 44% to 74% ) to discriminate active rejection from
no rejection. This is graphed as the sensitivity and specificity
over the range of dd-cfDNA (Figure 3E). The range of PPVand
NPV for dd-cfDNA for discriminating active rejection is
shown in Fi gure 3F; the PPV was 61% and NPV was 84%,
with the 1.0% dd-cfDNA cutoff.
Serum creatinine at time of biopsy did not discriminate
active rejection from no active rejection (Figure 3B). The
ROC cur ve for creatinine to discr iminate active rejection
had an AUC of 0.54 (95% CI, 0.43 to 0.66); i.e., at any cut-
off level for creatinine, there were as many false as true positive
resul ts (F igure 3D).
When the cohort of ABMR (including mixed ABMR and
TCMR) was compared with the cohort of all non-ABMR (in-
cluding TCMR-only), the fraction of dd-cfDNA differed
significantly ( P,0.001, Figure 5A), whereas there wa s no dis-
crimin ation by serum creatinine (Figure 5B). The fraction of
true positive results and the fraction of false positive res ults for
dd-cfDNA to discriminate ABMR status are shown in Figure
5C. The AUC was 0.87 (95% CI, 0.75 to 0.97). With a cutoff of
1.0%, dd-cfDNA has an 83% specificity (95% CI, 78% to 89%)
and 81% sensitivity (95% CI, 67% to 100%) to dis criminate
ABMR from no ABMR. The sensitivity and specificit y to
Figure 1. Banff elementary lesions and clinical features correlat e with dd-cfDNA level. The 107 samples (27 patients with 27 samples
with active rejection; 75 patients with 80 samp les with no active rejection) are rank-ordered and color-coded by dd -cfDNA level. White
indicates the element was not associated with that biopsy/visit. For each sample (x axis), associated elements (y axis) are shown as a
color ed box, by the level of dd-cf DNA associated with the sample; hi ghest dd-cfDNA in red, lowest in blue, with a vertical dashed line
at the 1% cutoff. The si gnificance (P value) of association of dd-cfDNA .1% with each element is shown. BM, (glomerular) basement
membrane; CNI, calcineurin inhibitor; DGF, delayed graft function; ENDATs, (gene expression profiles of) endoth elial activation (and
injury) transcripts; Inflam, inflammation; ptc, peritubular capillary.
J Am Soc Nephrol 28: 2221–2232, 2017 Cell-free DNA and Kidney Rejection 2223
www.jasn.org
CLINICAL RESEARCH
discriminate ABMR over the range of potential cutoffs is
shown in Figure 5E. The range of PPV and NPV for discrim-
inating ABMR is shown in Figure 5F; the PP V was 44% and
NPV was 96% w ith the 1.0% dd-cfDNA cutoff. The ROC
curve for creatinine to discriminate ABMR h ad an AUC of
0.57 (95% CI, 0.42 to 0.71) (Figure 5D).
Among the 58 active rejections found in the clinically in-
dicated biopsies, the available 2 7 paired dd-cfDNA results are
shown in Figure 4A, broken out by rejection subclass: ten were
chronic, active ABMR; six acute/active ABMR; and 11 TCMR
only (types IA [5], IB [5], and IIA [1 ]). A s shown, the lowest
types of TCMR (type IA) had lower dd-cfDNA than t y pe IB or
type IIA (Figure 4B), although the number of cases was very
limited. The similarity in the pattern of dd-cfDNA values in
the nom inal two classes of ABMR was not surprising, because
the histologi c criteria overlap for these form s of ABMR. Figure
6 shows the dd-cfDNA results in the same 27 cases of active
rejection, categorized by other findings in addition to histo-
logic evidence of active rejection. For comparison, Figure 7
shows the results in the 80 biopsy specimens with no rejection,
categorized by other histolog ic findings. In both the active
rejection and no active rejection groups, interstitial fibrosis/
tubular atrophy (IF/TA) and acute tubular necrosis were rel-
atively common coincidental findings, an d no obvious trend
in the dd-cfDNA was associated with these. Because of the
small number of cases of these other diagnoses, including cal-
cineurin in hibitor (CNI) toxicity and BK virus (BKV), statis-
tical analyse s of these patterns were not performed.
Of 80 clinically indicated biopsies with no act ive re jection
find ings, only ni ne biopsy specimens were reported to show
essentially normal histology (i.e., no other coincidental find-
ings, such as IF/TA, acute tubular necrosis, BKV, GN, CNI
toxicity). We performed a comparison of dd-cfDNA in the
group of the normal biopsy speci mens to the biopsy specimens
showing no active rejection but one or mo re coincidental find-
ings: in the normal group (n=9) the median dd-cfDNA was
0.53% (interquarti le range, 0.22%–0.67%); in the coinc iden-
tal finding group (n=71), the median dd-cfDNA was
0.30% (interquartile ran ge, 0.14%–0.77%) (Wilcoxon ran k
sum test P=0.9).
Among the 107 biopsy specimens in either the active re-
jection or no active rejection groups, there were two reports in
Figure 2. Patients, blood samples, and biopsies used in this study.
2224 Journal of the American Society of Nephrol ogy J Am Soc Nephrol 28: 2221 –2232, 2017
CLINICAL RESEARCH
www.jasn.org
which the patho logists noted findings of papillomaBKvirus. In
case one, there was a viral load of .2 mill ion copies of BKV/ml
blood, and inflam mation equivalent to Banff 1B intensity (i2–
3, t3). In case two, there was moderate IF/TA and 9.99 million
copies of BKV/ml bl ood. The dd-cfDNA level was 4.6% and
2.3%, respectively, in these cases.
DISCUSSION
In this study, mos t (204 of 242) kidney transplant biopsies were
triggere d by an elevation in serum cre atinine over baseline with
concerns for al loimmune injury, yet only 27% of these clini-
cally indica ted biopsies revealed active reje ction. The results in
the 107 biop sy specimens paired with plasma cfDNA showed
that dd-cfDNA levels di scriminated an active reject ion status
with an ROC-AUC of 0.74 and provided an estimated NPV
84% and PPV 61 % at a cutoff of 1.0% dd-cfDNA. These results
validated and extended prior reports of th e performance char-
acteristics of this assay.
7
The re was stronger performance of
dd-cfDNA in discriminating AB MR from no ABMR allograft
status (ROC-AUC 0.87, NPV 96%, PPV 44%, cutoff of 1.0%
dd-cfDNA). dd-cfDNA in 16 cases o f ABMR was significantly
higher (2.9%) than in 11 cases of TCMR rejection (0.2% in
type IA [five patients], 1.2% i n combined type IB [five pa-
tients] an d type IIA [one patient ]), and 0. 3% in the no active
rejection cohort (n=80). Because there is a clear trend that dd-
cfD NA was higher in type IB TCMR than in type IA, we spec-
ulate that dd-c fDNA i s li kely to be higher in the more severe
types of TCMR, but this cohor t did not have enough cases to
test this hypothesis.
The elevation of dd-cfDNA (.1%) was significantly asso-
ciated with ac ute/ active and chronic, active ABMR (Figure 1).
The dd-cfDNA levels across the threshold of 1% also
Table 1. Patient characteristics
Clinical Characteristic Active Rejection Group No Active Rejection Group P Value
a
Number of pati ents 27 75
Number of samples 27 80
Race, n (%) 0.23
Black 13 (48) 23 (31)
White 13 (48) 41 (55)
Native Hawaiian or Oth er Pacific Islander 1 (4) 0 (0)
Hispanic/Latino 0 (0) 4 (5)
Asian 0 (0) 1 (1)
Other 0 (0) 6 (8)
Men, n (%) 16(59) 45(60) .0.99
Age at enrollment, y 46616 53613 0.04
Post-transplant, d 96861107 118961482 0.42
CMV serologic status, n (%) 0.15
D2/R+ 4 (15) 13 (17)
D+/R+ 5 (19) 24 (32)
D2/R2 3(11) 16(21)
D+/R2 4(15) 9(12)
Unknown 11 (41) 13 (17)
Donor ty pe, n (%) 0.03
Deceased donor 20 (74) 42 (56)
Living unrelated 2 (7) 24 (32)
Living related 5 (19) 9 (12)
Child 2 (7) 3 (4)
Sibling 2 (7) 4 (5)
Parent 0 (0) 1 (1)
Half-sibling 0 (0) 0 (0)
Other biologic blood relation 1 (4) 1 (1)
Creatinine 2.561.0 2.461.4 0.69
eGFR 32612 36621 0.21
HLA class 1 no. of mismatches (A, B) 2.761.4 2.661.4 0.59
HLA class 2 no. of mismatches (DR) 1.260.6 1.160.8 0.67
Weight, kg 85619 84621 0.73
Height, cm 170610 171680.58
Data ranges are presented as mean6standard deviation. CMV, cytomegalovirus.
a
The P values are the level of statistical significa nce in the dif ferences of value s fou nd in the D ART active rejec tion gr oup and the no active rejection group.
For continuous covaria tes, Wilcoxon rank sum test was u sed to generate the P values. For categoric covariates, Fisher exact test was used to generate the
P va lues.
J Am Soc Nephrol 28: 2221–2232, 2017 Cell-free DNA and Kidney Rejection 2225
www.jasn.org
CLINICAL RESEARCH