j
`
`g
`
`y g
`
`p
`
`y
`
`Regular Article
`
`TRANSPLANTATION
`
`Plasma CXCL9 elevations correlate with chronic GVHD diagnosis
`Carrie L. Kitko,1 John E. Levine,1 Barry E. Storer,2 Xiaoyu Chai,2 David A. Fox,3 Thomas M. Braun,4 Daniel R. Couriel,1
`Paul J. Martin,2 Mary E. Flowers,2 John A. Hansen,2 Lawrence Chang,1 Megan Conlon,1 Bryan J. Fiema,1 Rachel Morgan,3
`Prae Pongtornpipat,1 Kelly Lamiman,1 James L. M. Ferrara,1 Stephanie J. Lee,2 and Sophie Paczesny5
`
`1Blood and Marrow Transplant Program, University of Michigan Comprehensive Cancer Center, Ann Arbor, MI; 2Fred Hutchinson Cancer Research Center,
`Seattle, WA; 3Division of Rheumatology, University of Michigan Health System, Ann Arbor, MI; 4Department of Biostatistics, University of Michigan, Ann
`Arbor, MI; and 5Bone Marrow and Stem Cell Transplantation Program, Indiana University Melvin and Bren Simon Cancer Center and Wells Center for
`Pediatric Research, Indianapolis, IN
`
`Key Points
`
`• Plasma concentrations of
`CXCL9 are elevated at the
`onset of cGVHD diagnosis,
`but not in patients with cGVHD
`for more than 3 months.
`• Plasma concentrations of
`CXCL9 are impacted by
`immunosuppressive therapy.
`
`There are no validated biomarkers for chronic GVHD (cGVHD). We used a protein mi-
`croarray and subsequent sequential enzyme-linked immunosorbent assay to compare 17
`patients with treatment-refractory de novo–onset cGVHD and 18 time-matched control
`patients without acute or chronic GVHD to identify 5 candidate proteins that distinguished
`cGVHD from no cGVHD: CXCL9, IL2Ra, elafin, CD13, and BAFF. We then assessed the
`discriminatory value of each protein individually and in composite panels in a validation
`cohort (n 5 109). CXCL9 was found to have the highest discriminatory value with an area under
`the receiver operating characteristic curve of 0.83 (95% confidence interval, 0.74-0.91). CXCL9
`plasma concentrations above the median were associated with a higher frequency of cGVHD
`even after adjustment for other factors related to developing cGVHD including age, diagnosis,
`donor source, and degree of HLA matching (71% vs 20%; P < .001). A separate validation
`cohort from a different transplant center (n 5 211) confirmed that CXCL9 plasma
`concentrations above the median were associated with more frequent newly diagnosed cGVHD after adjusting for the aforementioned factors
`(84% vs 60%; P 5 .001). Our results confirm that CXCL9 is elevated in patients with newly diagnosed cGVHD. (Blood. 2014;123(5):786-793)
`
`Introduction
`
`Improvements in survival following allogeneic hematopoietic cell
`transplantation (HCT) have been achieved by decreasing early post-
`HCT toxicities through better HLA matching, improved supportive
`care, and less toxic conditioning regimens. Despite multiple clinical
`trials investigating innovative treatments for chronic graft-versus-
`host disease (cGVHD), standard treatment has not changed in the
`past 30 years and cGVHD remains the leading cause of morbidity
`and mortality for long-term transplant survivors.1 The reasons for this
`lack of improvement are multifactorial and include an incomplete un-
`derstanding of the pathophysiology as well as inconsistent definitions for
`diagnostic and response criteria. In 2005, the National Institutes of
`Health Consensus Development Project on Criteria for Clinical Trials in
`cGVHD published a series of articles to help standardize the clinical
`approach to these patients and promoted new interest in this important
`posttransplant complication.2,3
`Acute GVHD (aGVHD) biomarkers have been identified that
`predict disease occurrence, distinguish new-onset GVHD from non-
`GVHD, have organ specificity, and can predict treatment response.4-8
`There is increasing interest in identifying cGVHD biomarkers that could
`also provide clinically meaningful information. Several publications
`have reported discovery of cGVHD biomarkers, but validation studies
`of biomarkers in independent populations are currently lacking.9-12
`Furthermore, newly diagnosed and established cGVHD cases are often
`studied together, although the pathologic processes culminating in
`
`a new diagnosis may be different than those present in established
`disease. Therefore, we focused on identifying biomarkers for newly
`diagnosed cGVHD. We interrogated patient samples with a microarray
`approach to identify candidate proteins elevated in the plasma of
`patients with newly diagnosed cGVHD. The leading 5 protein candi-
`dates were tested in 2 independent populations to validate the findings
`using high-throughput assays.
`Of the 5 proteins, chemokine (C-X-C motif) ligand 9 (CXCL9) had
`the most significant association with cGVHD. CXCL9 is an interferon-
`g–inducible ligand for chemokine (C-X-C motif) receptor 3 (CXCR3),
`which is expressed on effector CD41 Th1 cells and CD81 cytotoxic
`T lymphocytes. CXCL9 has been shown to influence the interactions
`and migration patterns of effector T cells to inflamed tissue.13 We
`found that CXCL9 was elevated in the plasma of all 3 cohorts studied
`and emerged as the best potential cGVHD biomarker.
`
`Methods
`
`Patients
`
`This study was approved by the institutional review boards (IRBs) of both the
`University of Michigan (UM) and the Fred Hutchinson Cancer Research
`Center (FHCRC). Informed consent was obtained from all patients or their
`
`Submitted August 9, 2013; accepted December 12, 2013. Prepublished online
`as Blood First Edition paper, December 20, 2013; DOI 10.1182/blood-2013-
`08-520072.
`
`The publication costs of this article were defrayed in part by page charge
`payment. Therefore, and solely to indicate this fact, this article is hereby
`marked “advertisement” in accordance with 18 USC section 1734.
`
`The online version of this article contains a data supplement.
`
`© 2014 by The American Society of Hematology
`
`786
`
`BLOOD, 30 JANUARY 2014 x VOLUME 123, NUMBER 5
`
`This material may be protected by Copyright law (Title 17 U.S. Code)
`
`

`

`j
`
`g
`
`y g
`
`p
`
`y
`
`BLOOD, 30 JANUARY 2014 x VOLUME 123, NUMBER 5
`
`PLASMA CXCL9 IS ELEVATED AT CHRONIC GVHD DIAGNOSIS
`
`787
`
`Table 1. Patient characteristics of the UM discovery and validation sets and FHCRC validation set
`
`UM discovery cohort (n 5 35)
`
`UM validation cohort (n 5 109)
`
`FHCRC validation cohort (n 5 211)
`
`No
`cGVHD
`(n 5 18)
`
`De novo
`cGVHD
`(n 5 17)
`
`No
`cGVHD
`(n 5 64)
`
`De novo
`cGVHD
`(n 5 45)
`
`P value
`
`P value
`
`P value
`(difference
`between
`discovery
`and
`validation)
`
`No
`cGVHD
`(n 5 33)
`
`New-onset
`cGVHD
`(n 5 86)
`
`Established
`cGVHD
`(n 5 92)
`
`P value
`(difference
`between no
`cGVHD and
`new-onset
`cGVHD)
`
`26
`
`0-66
`
`45
`
`19-58
`
`.12
`
`,.05
`
`41
`
`0–67
`
`50
`
`10–67
`
`.03
`
`.32
`
`54
`
`22-72
`
`52
`
`19-79
`
`52
`
`19-74
`
`.75
`
`53 (83%)
`
`44 (98%)
`
`.01
`
`.43
`
`33 (100)
`
`84 (98)
`
`88 (96)
`
`.38
`
`Characteristic
`
`Age, y
`
`Median
`
`Range
`
`Diagnosis
`
`Malignant*
`
`13 (72%)
`
`17 (100%)
`
`Nonmalignant†
`
`5 (28%)
`
`0
`
`11 (17%)
`
`1 (2%)
`
`0
`
`2 (2)
`
`4 (4)
`
`Disease status at HCT‡
`
`Low
`
`Intermediate
`
`High
`
`Donor type
`
`4 (31%)
`
`7 (54%)
`
`2 (15%)
`
`6 (35%)
`
`6 (35%)
`
`5 (30%)
`
`.65
`
`22 (42%)
`
`19 (43%)
`
`.71
`
`—
`
`17 (32%)
`
`11 (25%)
`
`14 (26%)
`
`14 (32%)
`
`11 (33)
`
`12 (36)
`
`10 (30)
`
`Matched sibling
`
`13 (72%)
`
`12 (71%)
`
`1.0
`
`40 (63%)
`
`25 (56%)
`
`.55
`
`.81
`
`18 (55)
`
`38 (44)
`
`30 (35)
`
`18 (21)
`
`30 (35)
`
`56 (65)
`
`33 (36)
`
`29 (32)
`
`29 (32)
`
`36 (39)
`
`56 (61)
`
`.45
`
`.05
`
`Other
`
`Source
`
`Bone marrow
`
`Cord blood
`
`Peripheral blood
`
`10 (56%)
`
`14 (82%)
`
`Conditioning intensity
`
`0
`
`0
`
`5 (28%)
`
`5 (29%)
`
`24 (37%)
`
`20 (44%)
`
`8 (44%)
`
`3 (18%)
`
`.15
`
`18 (28%)
`
`6 (13%)
`
`.10
`
`.69
`
`3 (5%)
`
`1 (3%)
`
`43 (67%)
`
`38 (84%)
`
`Full
`
`15 (83%)
`
`12 (71%)
`
`.44
`
`48 (75%)
`
`33 (73%)
`
`1.0
`
`.49
`
`3 (17%)
`
`5 (29%)
`
`16 (25%)
`
`12 (27%)
`
`15 (45)
`
`7 (21)
`
`0
`
`4 (5)
`
`4 (5)
`
`9 (10)
`
`0
`
`26 (79)
`
`78 (91)
`
`83 (90)
`
`19 (58)
`
`14 (42)
`
`50 (58)
`
`36 (42)
`
`55 (60)
`
`37 (40)
`
`.01
`
`.96
`
`Reduced
`
`aGVHD
`
`0
`
`I-IV
`
`NIH global severity
`
`Mild
`
`Moderate
`
`Severe
`
`18 (100%)
`
`17 (100%)
`
`39 (61%)
`
`45 (100%)
`
`—
`
`—
`
`0
`
`—
`
`—
`
`—
`
`0
`
`—
`
`6 (35%)
`
`11 (59%)
`
`25 (39%)
`
`0
`
`—
`
`—
`
`—
`
`—
`
`4 (9%)
`
`25 (56%)
`
`16 (35%)
`
`—
`
`.04
`
`9 (27)
`
`24 (73)
`
`—
`
`—
`
`—
`
`16 (19)
`
`70 (81)
`
`3 (3)
`
`46 (53)
`
`37 (43)
`
`27 (29)
`
`65 (71)
`
`10 (11)
`
`57 (62)
`
`25 (27)
`
`.30
`
`—
`
`*Malignant diseases included acute lymphoblastic leukemia, acute myelogenous leukemia, chronic lymphocytic leukemia, chronic myelogenous leukemia, chronic
`myelomonocytic leukemia, Hodgkin lymphoma, juvenile myelomonocytic leukemia, Kostmann syndrome, non-Hodgkin lymphoma, multiple myeloma, myelodysplastic
`syndrome, myeloproliferative disorder, paroxysmal nocturnal hematuria, and prolymphocytic leukemia.
`†Nonmalignant disease included malignant infantile osteopetrosis, severe aplastic anemia, sickle cell anemia, thalassemia, severe combined immunodeficiency disorder,
`X-linked lymphoproliferative disorder.
`‡Low-, intermediate-, or high-risk disease status according to Center for International Blood and Marrow Transplant Research guidance.
`
`legal guardians in accordance with the Declaration of Helsinki. Patient
`characteristics are summarized in Table 1. The UM discovery cohort consisted
`of 17 patients with treatment refractory de novo–onset cGVHD (defined as
`rapidly progressive in severity or refractory to initial therapy) and 18 patients
`without a history of either aGVHD or cGVHD in order to identify 2 groups most
`likely to show differences in protein concentrations and to remove biomarkers
`only associated with aGVHD. The UM validation set was made up of a separate
`group of 109 patients. There were 45 patients with de novo–onset cGVHD who
`had prospectively collected plasma samples obtained within 50 days of the onset
`of cGVHD. There were an additional 64 patients who had plasma samples
`collected at matched time points to the 45 cGVHD patients but had not
`developed cGVHD at the time of sample acquisition and any aGVHD had
`resolved (22%). Both the UM discovery and validation patients provided plasma
`samples for an IRB-approved biorepository from 2002 to 2008. cGVHD-
`specific data were retrospectively reviewed by 2 clinicians (C.L.K. and
`D.R.C.) with expertise in cGVHD who confirmed that patients met the
`National Institutes of Health (NIH) consensus criteria for diagnosis of the
`disease and assigned individual organ involvement and global score according
`to the 2005 NIH Consensus Criteria.2 Details of cGVHD characteristics are
`provided in supplemental Table 1, available on the Blood Web site.
`A second independent validation set was composed of 211 patients treated
`at FHCRC from 2008 to 2011. The FHCRC validation cohort included samples
`obtained at the time of enrollment on an IRB-approved long-term follow-up
`study. Patients entered this study from 3 months to 66 months posttransplant;
`thus, there was greater heterogeneity in timing of sample acquisition relative
`to cGVHD onset. Therefore, we divided the FHCRC cohort into 3 groups:
`
`controls without cGVHD, newly diagnosed cGVHD (sample obtained within
`90 days of diagnosis), and those with established cGVHD (sample obtained
`3-36 months post-cGVHD diagnosis). Time to sample acquisition relative
`to HCT and diagnosis of cGVHD for both cohorts are provided in Table 2.
`In contrast to the UM patients, the FHCRC cGVHD cohort included all types
`of cGVHD presentation (de novo, quiescent, and progressive). In both the
`UM and FHCRC cohorts, the onset of cGVHD was defined as the first time the
`NIH consensus criteria for diagnosis of cGVHD occurred,2 which was not
`necessarily when a patient first received systemic therapy.
`
`Antibody array and ELISA
`
`Plasma samples in the discovery set were analyzed using a customized quantitative
`microarray dotted with 130 antibodies that targeted a diverse group of proteins
`detailed in supplemental Table 2 (RayBiotech, Norcross, GA). Briefly, we used an
`array of matched-pair antibodies for detection of each target protein. Samples
`(50 mL) were incubated with the arrays, nonspecific proteins were washed off,
`and detection was carried out using a cocktail of biotinylated antibodies,
`followed by a streptavidin-conjugated fluor. Signals were visualized using
`a fluorescence laser scanner and quantified by comparison with array-specific
`protein standard curves. Proteins that could distinguish between the cGVHD-
`positive and cGVHD-negative groups with a P value # .1 met the threshold for
`validation with enzyme-linked immunosorbent assay (ELISA).
`Validation of the proteins of interest from the microarray was performed
`with a sequential ELISA protocol to maximize the number of measured
`analytes per sample by reusing the same aliquot consecutively in individual
`
`

`

`j
`
`g
`
`y g
`
`p
`
`y
`
`788
`
`KITKO et al
`
`BLOOD, 30 JANUARY 2014 x VOLUME 123, NUMBER 5
`
`Table 2. Time to sample acquisition for UM and FHCRC cohorts
`
`UM discovery cohort (n 5 35)
`
`UM validation cohort (n 5 109)
`
`FHCRC validation cohort (n 5 211)
`
`No
`cGVHD
`(n 5 18)
`
`De novo
`cGVHD
`(n 5 17)
`
`No
`cGVHD
`(n 5 64)
`
`De novo
`cGVHD
`(n 5 45)
`
`P value
`
`P value
`
`P value*
`
`No
`cGVHD
`(n 5 33)
`
`New-onset
`cGVHD
`(n 5 86)
`
`Established
`cGVHD
`(n 5 92)
`
`P value†
`
`102
`
`94-189
`
`103
`
`97-221
`
`.25
`
`135
`
`92-205
`
`154
`
`55-364
`
`0.25
`
`369
`
`256
`
`619
`
`,.001
`
`161-3641
`
`92-915
`
`196-8974
`
`.14
`
`.45
`
`—
`
`—
`
`0
`250 to 112
`
`—
`
`—
`
`—
`
`0
`242 to 135
`
`—
`
`—
`
`—
`
`11
`
`0-91
`
`389
`
`92-1168
`
`,.001‡
`
`Characteristic
`
`Time post-HCT,
`
`days
`
`Median
`
`Range
`
`Time post-cGVHD
`
`onset, days
`
`Median
`
`Range
`
`*Difference between discovery and validation.
`†Difference between No cGVHD and new-onset cGVHD.
`‡Difference between new onset vs established.
`
`ELISA plates. Commercial antibody pairs were available for CXCL9
`(RayBiotech), elafin, interleukin 2 receptor a (IL2Ra), and soluble B-cell–
`activating factor (BAFF) (R&D Systems, Minneapolis, MN). The specificity
`of the capture and detection antibodies for CXCL9 from RayBiotech is as
`follows. For capture antibody: host, mouse; isotype, mouse immunoglobulin
`G1; k, immunogen, baculovirus-expressed full-length recombinant human
`CXCL9 protein; clonality, monoclonal. For detection antibody: host, mouse;
`isotype, mouse immunoglobulin G1; k, immunogen, baculovirus-expressed
`full-length recombinant human CXCL9 protein; Clonality: Monoclonal.
`These antibodies have shown ,0.1% cross-reactivity with many human CXC
`chemokines (CXCL1, CXCL2, CXCL3, CXCL4/PF4, CXCL7, and CXCL10)
`as well as a variety of other immunologic proteins. Samples and standards were
`analyzed in duplicate according to a previously described protocol.14
`In addition, because CD13 has been reported to be elevated in patients at
`onset of cGVHD,11 we developed a novel sandwich ELISA using 2 mouse
`anti–human CD13 monoclonal antibodies directed at distinct epitopes of
`CD13 to analyze CD13 plasma concentrations in the discovery set. Briefly,
`plates were coated with anti-CD13 antibody WM1515 in carbonate buffer and
`then blocked with a blocking solution devoid of animal protein (Vector
`Laboratories, Burlingame, CA). Test samples were applied and CD13 was
`detected using a biotinylated anti-CD13 antibody termed 591.1D7.34 that
`was generated in the Fox laboratory, followed by streptavidin/horseradish
`peroxidase and TMB substrate. We used the same technique for measuring
`CD13 concentrations in the validation cohort as CD13 met our a priori criteria
`for a candidate biomarker. Plasma samples were run by a technician blinded to
`clinical factors or case/control status.
`
`Statistical methods
`
`Differences in the groups with and without cGVHD were compared with
`Student t tests for continuous variables and with Fisher’s exact tests for
`categorical variables. Differences in patient characteristics between training
`and validation sets were assessed with a Breslow-Day test for homogeneity
`of the odds ratios. Median protein concentrations were compared using the
`Wilcoxon-Mann-Whitney test. The x2 test was used for unadjusted comparison
`of proportions. Logistic regression with adjustment for clinical factors
`known to be related to cGVHD in the 2 cohorts was used to compare pro-
`portions of patients with cGVHD in the high vs low CXCL9 groups,
`classified by division at the median. A probability level of ,.05 was consid-
`ered to be statistically significant. P values were not corrected for multiple
`comparisons in a priori analyses. Receiver operating characteristic (ROC) area
`under the curves (AUC) were estimated nonparametrically.
`
`Results
`
`We hypothesized that samples at onset of de novo cGVHD from
`patients who ultimately developed treatment-refractory disease would
`
`be most likely to contain cGVHD-specific biomarkers. Using the
`protein microarray (supplemental Table 2) and subsequent ELISA
`workflow outlined, we identified 5 proteins (out of the 131 tested;
`130 from the microarray 1 CD13, which was measured separately)
`that distinguished refractory cGVHD patients at disease onset from
`patients who never had aGVHD or cGVHD: CXCL9, IL2Ra, elafin,
`CD13, and BAFF (Figure 1A-E).
`We then measured concentrations of these 5 proteins in samples
`from the validation cohort of UM patients. Of note, patients in the
`cGVHD group were older and more likely to have received
`a transplant for a malignant condition than the no-cGVHD controls.
`Otherwise, there were no statistically significant differences between
`the patients with cGVHD and without cGVHD based on donor type,
`graft source, HLA match, or conditioning intensity. Likewise, samples
`were collected at similar times for both the cGVHD cases and controls.
`Samples were obtained at a median of 154 days after HCT in the
`cGVHD group compared with 135 days after HCT in the no-cGVHD
`group (P 5 .25). As in the discovery set, all 5 candidate proteins were
`significantly elevated in patients with newly diagnosed de novo–onset
`cGVHD compared with those without cGVHD (Figure 1F-J), vali-
`dating our initial findings. As others have also reported, we found an
`association of higher CD13 concentrations in patients whose cGVHD
`included liver involvement compared with cGVHD patients without
`liver involvement (median 1382 vs 725 ng/mL; P , .0001).9
`To better define the potential clinical utility of these proteins
`elevated at the onset of cGVHD, we performed area under the ROC
`curve analyses for each protein comparing no cGVHD to de
`novo–onset cGVHD. The AUCs were similar for IL2Ra, elafin,
`CD13, and BAFF and ranged from 0.62–0.67 while the AUC for
`CXCL9 was 0.83 (supplemental Figure 1A). Given the similar AUCs
`for 4 of the proteins, we combined them into a composite panel
`(without CXCL9), which provided an improved AUC of 0.74. When
`CXCL9 was added to the composite panel, the AUC improved
`further to 0.83 but was not better than CXCL9 alone (supplemental
`Figure 1B). Because there was no additional diagnostic value to
`using the composite panels, we determined that CXCL9 had the best
`correlation with de novo–onset cGVHD and further analyses were
`confined to CXCL9.
`Next, we determined that the median CXCL9 plasma concentra-
`tion provide an 87% sensitivity and a 77% specificity for identifying
`de novo cGVHD (supplemental Table 3). We then assessed the cor-
`relation of CXCL9 plasma concentrations and diagnosis of cGVHD
`by x2 analysis. CXCL9 plasma concentrations above the median
`(6.5 pg/mL) were strongly associated with the presence of newly
`diagnosed cGVHD (71% vs 20%; P , .001), a finding that remained
`
`

`

`j
`
`g
`
`y g
`
`p
`
`y
`
`BLOOD, 30 JANUARY 2014 x VOLUME 123, NUMBER 5
`
`PLASMA CXCL9 IS ELEVATED AT CHRONIC GVHD DIAGNOSIS
`
`789
`
`A
`
`Discovery
`
`F
`
`Figure 1. Biomarkers at onset of cGVHD. (A-E) ELISA
`results of median plasma concentrations of CXCL9
`(A), BAFF (B), CD13 (C), IL2Ra (D), and elafin (E) in
`the no cGVHD patients (n 5 18) and refractory de novo
`cGVHD patients (n 5 17) from the discovery cohort.
`(F-J) ELISA results of median plasma concentrations
`of CXCL9 (F), BAFF (G), CD13 (H), IL2Ra (I), and
`elafin (J) in the non-cGVHD patients (n 5 64) and de
`novo cGVHD patients (n 5 45) from the validation
`cohort. Data are illustrated as box and whisker plots with
`the whiskers indicating the 90th and 10th percentiles.
`
`80
`
`......
`e 60
`c,
`.s
`
`0)
`
`40
`
`...J g
`
`~ 20
`
`(,)
`
`0
`
`NoGVHD
`
`cGVHD
`
`B 20000
`
`p=0.0001
`
`115000
`
`LL
`LL
`<C
`IXI
`
`5000
`
`0
`
`~ 10000 B
`
`NoGVHD
`
`cGVHD
`
`C 8000
`......
`~ 6000
`.s 4000
`....
`C 2000
`(,)
`
`C)
`
`I"')
`
`p=0.003
`
`I
`
`0
`
`NoGVHD
`
`cGVHD
`
`p=0.04
`
`NoGVHD
`
`cGVHD
`
`p=0.057
`
`Validation
`
`e<o.0001
`
`80
`
`60
`
`40
`
`......
`e
`c,
`.s
`
`(,)
`
`0)
`...J
`~ 20
`
`0
`
`G 15000
`
`...... i 10000
`
`LL
`~ 5000
`IXI
`
`0
`
`H 5000
`...... 4000
`e
`c, 3000
`.s
`....
`I"') 2000
`C
`(,) 1000
`
`0
`
`3000
`
`NoGVHD
`
`cGVHD
`
`s
`
`NoGVHD
`
`p=0.03
`
`cGVHD
`
`I
`
`NoGVHD
`
`cGVHD
`
`......
`
`0
`ix: 1000
`~
`
`0
`
`J 10000
`...... 8000
`I
`.s
`C 4000
`i
`iii 2000
`
`0
`
`NoGVHD
`
`cGVHD
`
`p=0.002
`
`D 4000
`w 3000
`
`.S 2000
`0
`ix:
`~ 1000
`
`0
`
`E 8000
`......
`E 6000
`c,
`.S 4000
`q::
`111 2000
`iii
`
`0
`
`C $
`
`I 2000 $
`c, ~ ~
`C) 6000 a
`
`NoGVHD
`
`cGVHD
`
`NoGVHD
`
`cGVHD
`
`statistically significant (P , .001) after adjusting for potential con-
`founding factors associated with the development of cGVHD
`(patient age, graft source [bone marrow/cord blood vs peripheral
`blood HCT], HLA match [matched sibling vs other] and diagnosis
`[malignant vs nonmalignant]) (Table 3).
`Finally, we assessed if CXCL9 concentrations were associated
`with other factors. Since changes in CXCL9 concentrations may
`reflect differences in immune recovery, we first analyzed for an as-
`sociation of CXCL9 concentrations and absolute lymphocyte count,
`and found none. We then examined whether CXCL9 concentrations
`were higher as time post-HCT increased, an alternative way to look
`for an association with CXCL9 and immune recovery. In the cGVHD
`
`patients, we did not detect an association of CXCL9 concentration
`and time post-HCT. Therefore, we concluded that CXCL9 elevated
`concentrations at the time of de novo cGVHD were due to the
`presence of the disease. We then sought to further validate CXCL9
`as a marker of cGVHD activity in a second, more heterogeneous,
`independent cohort.
`We obtained 211 samples from the FHCRC for validation. Unlike
`the UM cohort, the FHCRC validation cohort included patients with
`any type of cGVHD presentation (de novo, quiescent, or progressive).
`In order to create more homogenous subsets within the FHCRC
`cohort, we divided the cGVHD patients into a newly diagnosed
`group (within 90 days of diagnosis; n 5 86) and an established
`
`

`

`j
`
`g
`
`y g
`
`p
`
`y
`
`790
`
`KITKO et al
`
`BLOOD, 30 JANUARY 2014 x VOLUME 123, NUMBER 5
`
`Table 3. x2 association for CXCL9 levels above the median with
`cGVHD
`
`Total number per group (cGVHD %)
`
`Less than or
`equal to median
`
`More than
`median
`
`UM validation
`
`Newly diagnosed FHCRC
`
`Established FHCRC
`
`64 (20%)
`
`58 (60%)
`
`71 (68%)
`
`45 (71%)
`
`61 (84%)
`
`54 (81%)
`
`P value*
`
`,.001
`.001
`
`.04
`
`CI, confidence interval.
`*Adjusted for age, stem cell source (bone marrow/cord blood vs peripheral blood),
`HLA match (matched sibling vs other), and diagnosis (malignant vs nonmalignant).
`
`cGVHD group (diagnosed 3-36 months prior to sample acquisition;
`n 5 92). Control patients (n 5 33) did not have cGVHD, but prior
`treated aGVHD was allowed (73%). The median plasma concentra-
`tion of CXCL9 was significantly higher in the FHCRC cohort than in
`the UM cohort (26 vs 6.5 pg/mL; P , .0001), presumably reflecting
`the differences in the 2 populations described above. For our initial
`analysis of this independent cohort, we limited comparisons to no
`cGVHD controls vs newly diagnosed patients because they were
`most similar to the UM cohort. Despite differences in absolute values
`of CXCL9, as in the UM results, CXCL9 plasma concentrations were
`significantly higher in patients with newly diagnosed cGVHD
`compared with the no-cGVHD patients (P 5 .003; Figure 2). Area
`under the ROC curve analyses for CXCL9 comparing controls with
`no cGVHD with patients with newly diagnosed cGVHD revealed an
`AUC of 0.68 with a sensitivity and specificity at the median of 59%
`and 70%, respectively (supplemental Table 3). Given the similarity
`of these results to those seen in the UM validation set, we performed
`an identical adjusted x2 analysis for the FHCRC newly diagnosed
`cGVHD patients. As in the UM analysis, CXCL9 plasma
`concentrations above the median were strongly associated with the
`presence of cGVHD (84% vs 60%; P 5 .001; Table 3).
`Given the strong correlation between CXCL9 plasma concen-
`trations above the median and the presence of newly diagnosed
`cGVHD, we evaluated whether CXCL9 plasma concentrations were
`also associated with cGVHD severity at diagnosis. Very few patients
`in either the UM cohort (n 5 4) or the newly diagnosed FHCRC
`cohort (n 5 3) had mild cGVHD, so those patients were combined
`with patients who presented with moderate cGVHD. In both the UM
`cohort and FHCRC cohorts, CXCL9 plasma concentrations were
`significantly higher in patients who presented with severe cGVHD
`compared with no cGVHD group (P , .0001 and P 5 .0009 re-
`spectively; Figure 3A-B). Although UM patients who presented with
`mild/moderate cGVHD had significantly higher CXCL9 plasma
`concentrations compared with no cGVHD controls (P , .001;
`Figure 3A), we were unable to reproduce this finding in the FHCRC
`patients with mild/moderate cGVHD (P 5 .17; Figure 3B).
`Finally, because previously reported biomarkers for both acute
`and chronic GVHD have been shown to decrease following initiation
`of immunosuppressive therapy (IST),10,16 we analyzed the effect of
`treatment with IST on CXCL9 concentrations. In the UM cohort,
`where samples were obtained closer to the time of onset and possible
`initiation of therapy, we found that median CXCL9 concentrations
`were higher in patients not on IST (n 5 19) compared with patients
`on IST (n 5 43; 39 vs 15 ng/mL; P 5 .009); furthermore, both groups
`had higher concentrations than the no cGVHD controls (n 5 82,
`4 ng/mL; P , .001 for both comparisons; Figure 4A). We performed
`the same analysis in the newly diagnosed FHCRC cohort. As in the
`UM cohort, patients not on IST at the time of sample acquisition (n 5 43)
`had higher CXCL9 concentrations than patients on IST (n 5 43;
`77 vs 23 ng/mL; P , .0001; Figure 4B) and the no cGVHD controls
`
`(n 5 33; 20 ng/mL; P , .0001). Unlike the UM cohort however,
`concentrations of CXCL9 in patients on IST was not higher than the no
`cGVHD controls (P 5 .51). This result might be explained by
`differences in the intensity and duration of IST between the cohorts.
`UM patients on IST were generally not on systemic steroids at the
`time of sample acquisition (84%), whereas only 2% of FHCRC
`patients were not treated with steroids when samples were acquired.
`Taken together, this finding suggests that intensity and duration of
`cGVHD treatment lowers CXCL9 concentrations. Lastly, because
`both cohorts consisted entirely of patients with multiorgan involvement,
`we could not validate CXCL9 as a biomarker with target organ
`specificity (data not shown).
`We also were able to study CXCL9 concentrations in the FHCRC
`patients with established cGVHD (n 5 92; sample obtained 3-36
`months after cGVHD diagnosis). CXCL9 plasma concentrations in
`this group of patients with long-standing and treated cGVHD were
`not statistically different compared with the no-cGVHD controls
`(P 5 .18). Likewise, there was no correlation between CXCL9
`plasma concentrations above the median and the presence of cGVHD
`(Table 3) or by disease severity (data not shown).
`
`Discussion
`
`Discovery of valid and reproducible biomarkers for cGVHD remains
`a significant challenge. Compared with aGVHD, cGVHD is clinically
`more heterogeneous and can involve many more target organs, often
`simultaneously. Additionally, the timing of sample acquisition for
`biomarker assessment is also critical. Once immunosuppression
`has been initiated, the biomarker pattern may change, as has been
`previously been observed with BAFF plasma concentrations after
`patients are treated with corticosteroids10 and was observed in our
`study as well. Therefore, one of the strengths of our study design was
`the inclusion of only de novo cGVHD in the first validation cohort,
`when the length of prior therapy was minimized. Another strength
`of our study was that we were then able to reproduce the strong
`correlation of CXCL9 with cGVHD in a second more heterogeneous
`cohort. Taken together, these findings provide convincing evidence
`that elevated CXCL9 concentrations are a marker for newly
`diagnosed cGVHD.
`
`p=0.003
`
`c, 100
`
`150 -E
`C: -0)
`~50 g
`
`..J
`
`o~---r---------..----
`NocGVHD
`Newly Diagnosed
`
`n:
`
`33
`
`86
`Figure 2. CXCL9 is elevated in newly diagnosed cGVHD from an independent
`cohort. ELISA results of median plasma concentrations of CXCL9 from no cGVHD
`patients (n 5 33) and newly diagnosed cGVHD patients (n 5 86) in a second
`validation cohort from the FHCRC. Data are illustrated as box and whisker plots with
`the whiskers indicating the 90th and 10th percentiles.
`
`

`

`j
`
`g
`
`y g
`
`p
`
`y
`
`BLOOD, 30 JANUARY 2014 x VOLUME 123, NUMBER 5
`
`PLASMA CXCL9 IS ELEVATED AT CHRONIC GVHD DIAGNOSIS
`
`791
`
`Figure 3. Increased CXCL9 levels are associated
`with increased cGVHD severity. (A) ELISA results of
`median plasma concentration of CXCL9 from no GVHD
`(n 5 82), mild/moderate cGVHD (n 5 35), and severe
`cGVHD (n 5 27) from the entire UM cohort. (B) ELISA
`results of median plasma levels of CXCL9 from no
`GVHD (n 5 33), mild/moderate (mild/mod) cGVHD
`(n 5 49), and severe cGVHD (n 5 37) from the newly
`diagnosed FHCRC cohort. Data are illustrated as box
`and whisker plots with the whiskers indicating the 90th
`and 10th percentiles.
`
`A
`
`80 -E 60 -en
`
`C
`;- 40
`...I
`(.)
`>( 20
`(.)
`
`UM Cohort
`
`B
`
`FHCRC Cohort
`
`p<0.001
`
`p<0.0001
`
`e <o.0001
`e=0.0009
`
`p=0.17
`
`g
`
`-150
`
`en
`.:, 100
`a,
`...I
`
`>< (.)
`
`E -
`(.) 50 8
`
`0
`
`No cGVHD
`82
`
`n:
`
`Mild/Mod
`35
`
`Severe
`27
`
`NocGVHD Mild/Mod
`
`Severe
`
`n:
`
`33
`
`49
`
`37
`
`CXCL9 is an interferon-g–inducible chemokine that binds to
`CXCR3, its only known receptor. CXCR3 expression can be
`rapidly induced in both CD41 type 1 helper cells as well as CD81
`cytotoxic lymphocytes following dendritic cell activation of
`na¨ıve lymphocytes.13 In both human and mouse autoimmune disease
`studies, the binding of CXCL9 to CXCR3 promotes lymphocyte
`migration to inflamed tissues.17,18 CXCR3 has also been shown to be
`critical for the recruitment of alloreactive T cells in aGVHD,19,20
`whereas CXCL9 has been shown to be elevated in tissue samples
`from patients with oral,21 ocular,22 and cutaneous23 cGVHD. We
`found that the plasma of patients with newly diagnosed cGVHD, but
`not established cGVHD, contains higher concentrations of CXCL9
`than patients without cGVHD. These results suggest that this T-cell
`chemoattractant is involved in the initiating steps of the cGVHD
`disease process, particularly around the time that clinical manifes-
`tations are first noted. The role of CXCL9 in the pathophysiology of
`cGVHD after the disease is well established and systemic therapy has
`been given is not as clear. Given the well-described relationship of
`CXCL9-CXCR3 in Th1-mediated disease states, it is intriguing to
`speculate that the Th1 pathways may be important during the early
`stages of cGVHD.
`One other group reported that in a study of 28 patients with cGVHD,
`CXCL9 serum concentrations were associated with cGVHD involving
`the skin, but not other phenotypes.23 Our cohort did not include
`patients with isolated skin involvement, which precluded us from
`performing the same analysis. However, as noted, we did not find that
`CXCL9 correlated with any particular organ involvement (data not
`shown). Given our large sample size and reproducibility of our
`results in independent validation cohorts, we believe that CXCL9
`may be useful as a marker of cGVHD that presents with a variety
`of clinical phenotypes. Of note, the same group also reported an
`association of CXCL10 and CXCL11 and cGVHD.23 Though
`
`CXCL10 was in the discovery array, we did not find a difference in
`CXCL10 levels in our discovery experiments and CXCL11 was not
`included in our discovery array and, therefore, neither marker was
`pursued further.
`Several limitations should be noted. First, although we included
`a large number of candidate biomarkers in our discovery array, our
`approach was not unbiased in that we preselected the candidates for
`study. Thus, proteins not included in our array but associated with
`cGVHD were missed. In addition, a previous study demonstrated
`a strong correlation of BAFF/B-cell ratios with active cGVHD.11
`Our study did not include analyses such as B-cell enumeration, so we
`cannot confirm the BAFF/B-cell ratio correlation. Using plasma
`protein concentrations alone, we found that CXCL9 had the highest
`AUC and best sensitivity and specificity for diagnosis of cGVHD of
`the 5 proteins tested. A direct comparison of the diagnostic utility of
`CXCL9 compared with BAFF/B-cell ratios could be useful. Next, this
`study does not address whether CXCL9 can predict the development
`of cGVHD as our focus was on samples obtained at the time of
`diagnosis. Furthermore, although 2 independent validation cohorts
`were included, the similarities in HCT practice