Value of Donor–Specific Anti–HLA Antibody Monitoring and Characterization for Risk Stratification of Kidney Allograft Loss

Abstract

The diagnosis system for allograft loss lacks accurate individual risk stratification on the basis of donor–specific anti–HLA antibody (anti-HLA DSA) characterization. We investigated whether systematic monitoring of DSA with extensive characterization increases performance in predicting kidney allograft loss. This prospective study included 851 kidney recipients transplanted between 2008 and 2010 who were systematically screened for DSA at transplant, 1 and 2 years post-transplant, and the time of post–transplant clinical events. We assessed DSA characteristics and performed systematic allograft biopsies at the time of post–transplant serum evaluation. At transplant, 110 (12.9%) patients had DSAs; post-transplant screening identified 186 (21.9%) DSA-positive patients. Post–transplant DSA monitoring improved the prediction of allograft loss when added to a model that included traditional determinants of allograft loss (increase in c statistic from 0.67; 95% confidence interval [95% CI], 0.62 to 0.73 to 0.72; 95% CI, 0.67 to 0.77). Addition of DSA IgG3 positivity or C1q binding capacity increased discrimination performance of the traditional model at transplant and post-transplant. Compared with DSA mean fluorescence intensity, DSA IgG3 positivity and C1q binding capacity adequately reclassified patients at lower or higher risk for allograft loss at transplant (category–free net reclassification index, 1.30; 95% CI, 0.94 to 1.67; P<0.001 and 0.93; 95% CI, 0.49 to 1.36; P<0.001, respectively) and post-transplant (category–free net reclassification index, 1.33; 95% CI, 1.03 to 1.62; P<0.001 and 0.95; 95% CI, 0.62 to 1.28; P<0.001, respectively). Thus, pre– and post–transplant DSA monitoring and characterization may improve individual risk stratification for kidney allograft loss.

Donor–specific anti–HLA antibodies (anti-HLA DSAs) have been extensively reported to be strongly associated with increased risks of rejection and allograft loss.^1–6 Although their value for accurate risk stratification of transplant outcomes has not been determined in the current literature, the detrimental influence of anti-HLA DSAs on transplant outcomes has placed anti-HLA antibodies at the center of national and local allocation policies in the United States and Europe.^7–10

Today, anti-HLA DSAs are considered to be among the most important biomarkers for predicting allograft injury and loss. However, there is no consensus for defining their pathogenicity and no standard for their evaluation to guide clinical decision making.^11,12 The current conventional approach to pre– and post–transplant immunologic risk evaluation is on the basis of the assessment by sensitive techniques of anti–HLA antibody specificity and strength, most frequently expressed by the mean fluorescence intensity (MFI) provided by single–antigen flow bead techniques.^8–10,13

Recently, significant advances have occurred in our ability to diagnose patients with antibody-mediated rejection (ABMR) and link anti–HLA antibody characteristics to transplant outcomes. These advances include the assessment of the capacity of anti-HLA antibodies to bind complement, particularly C1q binding, and the characterization of their IgG subclass composition. Converging evidence has supported that the capacity of anti-HLA DSA to bind complement is associated with an increased risk of antibody-mediated injury and poor allograft survival in not only kidney transplant^14–20 but also, heart,^21,22 liver,²³ and lung²⁴ transplant. Furthermore, emerging data have emphasized the clinical relevance of the IgG subclass composition of anti-HLA DSAs and their relationships with allograft injury phenotype²⁵ and survival in kidney^25–27 and liver^23,28 transplantation.

Considering that one of the most pressing unmet needs in transplant medicine involves delineating the characteristics of circulating anti–HLA antibodies that confer pathogenesis and influence transplant outcomes, the Transplantation Society Antibody Consensus Group issued a call to action in 2013 and encouraged the transplant community to focus future efforts on clinical trials that include serial anti–HLA DSA monitoring with the assessment of anti–HLA DSA characteristics, including their complement binding capacity and IgG subclass composition.⁹ After decades of studies emphasizing the associations between anti-HLA antibodies and kidney transplant outcomes,²⁹ the key issue today is to evaluate whether systematic anti–HLA DSA monitoring integrating the assessment of antibody characteristics might improve risk stratification for allograft loss.³⁰ Stratifying patients by their immunologic risk has the potential to resolve the puzzle of alloimmune conditions determining allograft outcomes and increase long–term allograft and patient survival by improving the efficacy of allocation policies and therapeutic strategies.³¹

Our hypothesis was that systematic monitoring and precise characterization of anti-HLA DSAs, including their complement binding capacity and IgG subclass composition, might add to the predictive value for allograft loss of the conventional approach on the basis of their detection and strength assessed by MFI level. To test this hypothesis, we specifically designed a prospective study performed in a large and unselected population of kidney transplant recipients who underwent standardized monitoring of anti-HLA DSAs together with systematic allograft biopsies. We assessed the performance of prospective systematic monitoring and characterization of anti-HLA DSAs (MFI, C1q binding capacity, and IgG subclasses) to improve individual risk stratification for allograft loss.

Results

Patient Characteristics

This prospective study enrolled 851 patients among 906 consecutive recipients undergoing kidney transplantation between January 1, 2008 and December 31, 2010. The study flow chart is provided in Figure 1.

Figure 1.

Prospective post-transplant anti-HLA DSA screening using single-antigen Luminex technique identified 110/851 (12.9%) patients with circulating anti–HLA DSA at the time of transplantation and 186/851 (21.9%) patients with circulating anti-HLA DSA after transplantation. Tx, transplant.

The characteristics of the study population at the time of transplantation are summarized in Table 1. The median follow-up after transplantation was 5.3 years (interquartile range, 4.6–6.2).

Table 1.

Baseline characteristics of the study population

Characteristics	N
Recipient characteristics
Age, yr, mean±SD	851	50.3±12.8
Men, n (%)	851	512 (60.2)
Retransplantation, n (%)	851	143 (16.8)
Time since dialysis, y, mean±SD	733	4.8±4.5
Blood type, n (%)	851
A		386 (46.2)
B		69 (8.3)
O		352 (42.2)
AB		28 (3.4)
CKD, n (%)	851
Glomerulopathy		222 (26.1)
Vascular nephropathy		64 (7.5)
Chronic interstitial nephropathy		103 (12.1)
Malformative uropathy		161 (18.9)
Diabetes		84 (9.9)
Other		36 (4.2)
Not determined		181 (21.3)
Diabetes mellitus, n (%)	851	123 (14.5)
Body mass index, kg/m2, mean±SD	803	23.6±4.3
Donor characteristics
Age, yr, mean±SD	851	51.4±15.8
Men, n (%)	851	477 (56.05)
Type, n (%)	851
Living		156 (18.3)
Cerebrovascular death		370 (43.5)
Other cause of death		325 (38.2)
Diabetes mellitus, n (%)	775	47 (5.5)
Hypertension, n (%)	822	213 (25.0)
Body mass index, kg/m2, mean±SD	835	26.1±25.7
Terminal serum creatinine, μmol/L, mean±SD	851	87.1±52.7
Transplant characteristics
Cold ischemia time, h, mean±SD	851	17.1±9.8
PRA, n (%)	851
0%–20%		746 (87.7)
21%–50%		53 (6.2)
51%–80%		25 (2.9)
81%–100%		27 (3.2)
Calculated PRA, n (%)	851
0%–20%		430 (50.5)
21%–50%		129 (15.2)
51%–80%		266 (31.3)
81%–100%		26 (3.1)
HLA mismatch, mean±SD	851
A		1.0±0.7
B		1.2±0.7
DR		0.9±0.7
Anti-HLA DSA at the time of transplantation, n (%)	851	110 (12.9)

PRA, panel reactive antibody.

Anti–HLA DSA Characteristics According to Time of Detection

Anti–HLA DSA Characteristics at the Time of Transplantation

Among the 110 (12.9%) patients with circulating anti–HLA DSA at the time of transplantation, the DSA with the highest MFI level, the immunodominant donor–specific antibody (iDSA), was HLA class 1 in 52 (47.3%) patients and HLA class 2 in 58 (52.7%) patients, with a mean MFI of 5952.1±4213.4 and C1q binding capacity in 35 (31.8%) patients. IgG1 was positive for 82 (74.6%) iDSAs, IgG2 was positive for 48 (43.6%) iDSAs, IgG3 was positive for 31 (28.2%) iDSAs, and IgG4 was positive for 30 (27.3%) iDSAs.

Post–Transplant Anti–HLA DSA Characteristics

Among the 186 (21.9%) patients identified with anti-HLA DSA after transplantation, 86 (46.2%) patients were positive for anti-HLA DSA at the time of a clinical event, 55 (29.6%) patients were identified at 1 year after transplantation, and 45 (24.2%) patients were identified at 2 years after transplantation. In total, the iDSA was HLA class 1 in 76 (40.9%) patients and HLA class 2 in 110 (59.1%) patients, with a mean MFI of 5746.7±4627.7 and C1q binding capacity in 57 (30.7%) patients. IgG1 was positive for 137 (73.7%) iDSAs, IgG2 was positive for 80 (43.0%) iDSAs, IgG3 was positive for 42 (22.6%) iDSAs, and IgG4 was positive for 46 (24.7%) iDSAs.

The characteristics of post–transplant anti–HLA DSA at the time of detection are shown in Table 2.

Table 2.

Clinical, histologic and immunological characteristics according to the time of anti-HLA DSA detection

Characteristics	Time of Transplantation	All Post–Transplant DSA	Post–Transplant Clinical Event	1-yr Post-Transplant Screening	2-yr Post-Transplant Screening	P Valuea
n	110	186	86	55	45
Characteristics of anti-HLA DSAs
All anti–HLA DSAs
No., mean±SD	2.0±1.3	1.8±1.2	1.9±1.2	1.8±1.4	1.7±1.0	0.52
HLA class specificity, n (%)						0.59
1	32 (29.1)	44 (23.7)	23 (26.7)	12 (21.8)	9 (20.0)
2	32 (29.1)	82 (44.1)	33 (38.4)	25 (45.5)	24 (53.3)
1+2	46 (41.8)	60 (32.3)	30 (34.9)	18 (32.7)	12 (26.7)
iDSA
HLA class specificity, n (%)						0.04
1	52 (47.3)	76 (40.9)	43 (50.0)	16 (29.1)	17 (37.8)
2	58 (52.7)	110 (59.1)	43 (50.0)	39 (70.9)	28 (62.2)
Preexisting, n (%)	—	81 (43.6)	39 (45.4)	42 (76.4)	0	<0.001
MFI, mean±SD	5952.1±4213.4	5746.7±4627.7	5971.2±4941.5	6319.5±4972.2	4617.4±3273.5	0.41
C1q binding, n (%)	35 (31.8)	57 (30.7)	34 (39.5)	13 (23.6)	10 (22.2)	0.05
IgG subclasses, n (%)
IgG1	82 (74.6)	137 (73.7)	62 (72.1)	42 (76.4)	33 (73.3)	0.85
IgG2	48 (43.6)	80 (43.0)	37 (43.0)	23 (41.8)	20 (44.4)	0.97
IgG3	31 (28.2)	42 (22.6)	38 (44.2)	2 (3.6)	2 (4.4)	<0.001
IgG4	30 (27.3)	46 (24.7)	9 (10.5)	24 (43.6)	13 (28.9)	<0.001
Clinical and histologic characteristics
Clinical characteristics
eGFR at biopsy, ml/min per 1.73 m2, mean±SD	—	41.2±19.9	30.5±13.0	49.9±20.9	50.7±19.8	<0.001
Proteinuria, g/g, mean±SD	—	0.5±0.8	0.9±1.0	0.2±0.1	0.2±0.2	<0.001
Histologic characteristics
Acute/active ABMR, n (%)	—	102 (54.8)	52 (60.5)	31 (56.4)	19 (42.2)	0.13
Chronic/active ABMR, n (%)	—	30 (16.1)	13 (15.1)	10 (18.2)	7 (15.6)	0.88
TCMR, n (%)	—	17 (9.1)	11 (12.8)	2 (3.6)	4 (9.1)	0.18
g + ptc Score, mean±SD	—	2.4±1.8	2.9±1.9	2.1±1.4	1.9±1.7	0.004
i + t Score, mean±SD	—	1.1±1.8	1.4±2.1	0.6±1.2	1.1±1.9	0.19
v Score, mean±SD	—	0.2±0.6	0.3±0.7	0.1±0.3	0.1±0.5	0.012
cg Score, mean±SD	—	0.3±0.8	0.4±0.9	0.2±0.4	0.3±0.8	0.99
IF/TA score, mean±SD	—	1.2±1.0	0.9±0.8	1.3±1.0	1.5±1.0	0.001
cv Score, mean±SD	—	1.4±1.0	1.2±1.1	1.5±1.1	1.6±0.9	0.10
ah Score, mean±SD	—	0.8±0.8	0.7±0.8	0.9±0.9	1.0±0.9	0.07
C4d deposition, n (%)	—	59 (31.7)	42 (48.8)	11 (20.0)	6 (13.3)	<0.001

—, not applicable; TCMR, T cell–mediated rejection; g, glomerulitis; ptc, peritubular capillaritis; i, mononuclear cell interstitial inflammation; t, tubulitis; v, intimal arteritis; cg, allograft glomerulopathy; IF/TA, interstitial fibrosis/tubular atrophy; cv, vascular fibrous intimal thickening; ah, arteriolar hyaline thickening.

^aP values are for the comparisons of the patients with post–transplant anti–HLA DSA detected for clinical indication, at 1 year after transplantation, and at 2 years after transplantation.

A comparison of post–transplant anti–HLA iDSA characteristics according to their preformed/de novo status is shown in Supplemental Table 1.

Clinical and Histologic Characteristics at the Time of Post–Transplant Anti–HLA DSA Detection

All of the patients with post–transplant anti–HLA DSA (n=186) underwent kidney allograft biopsy at the time of anti–HLA DSA detection. Among them, 65 (34.9%) patients had clinical ABMR, and 67 (36.0%) patients had subclinical ABMR. The clinical and histologic characteristics, according to the time of post–transplant anti–HLA DSA detection, are provided in Table 2.

A comparison of clinical and histologic characteristics according to the preformed/de novo status of post–transplant anti–HLA iDSA is shown in Supplemental Table 2.

Conventional Determinants of Kidney Allograft Loss at the Time of Transplantation

The 5-year kidney allograft survival rate was 89.1% (95% confidence interval [95% CI], 86.7 to 91.1).

To build the conventional model for allograft loss at the time of transplantation, we considered all of the traditional recipient, donor, and transplant characteristics at the time of transplantation (Table 3, univariate analysis). The following independent predictors of allograft loss at the time of transplantation were identified: donor age (per 1-year increment; hazard ratio [HR], 1.02; 95% CI, 1.00 to 1.03; P=0.03), cold ischemia time (per 1-hour increment; HR, 1.03; 95% CI, 1.01 to 1.05; P=0.01), donor terminal serum creatinine (per 1-μmol/L increment; HR, 1.00; 95% CI, 1.00 to 1.01; P=0.01), and the presence of anti-HLA DSA at the time of transplantation (HR, 2.32; 95% CI, 1.40 to 3.84; P=0.001). These risk factors constituted the conventional model for allograft loss at the time of transplantation: the day 0 reference model (Table 3, multivariate day 0 reference model).

Table 3.

Conventional determinants of time to kidney allograft loss at the time of transplantation: univariate analysis and multivariate day 0 reference model

Variables	No. of Patients	No. of Events	HR	95% CI	P Value
Univariate analysis
Recipient characteristics
Age per 1-yr increment	851	86	0.99	0.98 to 1.01	0.46
Sex
Women	339	35	1	—
Men	512	51	0.95	0.62 to 1.46	0.80
Retransplantation
No	708	66	1	—
Yes	143	20	1.60	0.97 to 2.64	0.07
Time since dialysis per 1-yr increment	733	81	1.02	0.97 to 1.06	0.46
Diabetes mellitus
No	728	72	1	—
Yes	123	14	1.20	0.67 to 2.12	0.54
Body mass index per 1-kg/m2 increment	803	80	1.00	0.95 to 1.05	>0.99
Donor characteristics
Age per 1-yr increment	851	86	1.02	1.00 to 1.03	0.02
Sex
Women	374	39	1	—
Men	477	47	0.94	0.62 to 1.44	0.78
Type
Living	156	8	1	—
Deceased	695	78	2.29	1.10 to 4.73	0.03
Diabetes mellitus
No	728	73
Yes	47	5	1.17	0.47 to 2.89	0.74
Hypertension
No	608	56	1	—
Yes	214	28	1.52	0.97 to 2.40	0.07
Body mass index per 1-kg/m2 increment	835	84	1.00	0.99 to 1.01	0.88
Terminal SCr per 1-μmol/L increment	851	86	1.00	1.00 to 1.01	0.01
Transplant characteristics
Cold ischemia time per 1-h increment	851	86	1.04	1.02 to 1.06	<0.001
PRA per 10% increment	851	86	1.11	1.04 to 1.20	0.004
Calculated PRA per 10% increment	851	86	1.08	1.02 to 1.15	<0.01
HLA-A/-B/-DR mismatch (continuous)	851	86	0.90	0.79 to 1.04	0.16
Anti-HLA DSA
No	741	65	1	—
Yes	110	21	2.47	1.51 to 4.05	<0.001
Multivariate day 0 reference model
Donor age per 1-yr increment	851	86	1.02	1.00 to 1.03	0.03
Cold ischemia time per 1-h increment	851	86	1.03	1.01 to 1.05	0.01
Donor terminal SCr per 1-μmol/L increment	851	86	1.003	1.00 to 1.01	0.01
Anti-HLA DSA
No	741	65	1	—
Yes	110	21	2.32	1.40 to 3.84	0.001

—, not applicable; SCr, serum creatinine; PRA, panel reactive antibody.

Incremental Effect of Post–Transplant Anti–HLA DSA Monitoring on the Conventional Determinants of Allograft Loss at the Time of Transplantation

The addition of post–transplant anti–HLA DSA detected by single-antigen Luminex to the day 0 reference model (post–Tx DSA model) was significantly associated with the time to allograft loss (adjusted HR, 3.0; 95% CI, 1.8 to 5.0; P<0.001) and improved the ability to discriminate allograft loss, with an increase in the c statistic from 0.67 (95% CI, 0.62 to 0.73) for the day 0 reference model to 0.72 (95% CI, 0.67 to 0.77) for the post–Tx DSA model. The mean difference in the c statistic between the day 0 reference model and the post–Tx DSA model was 0.047 (95% CI, 0.046 to 0.049). The integrated discrimination improvement (IDI) between the post–Tx DSA model and the day 0 reference model was 0.021 (95% CI, <0.00 to 0.04; P=0.003). The detection of post–transplant anti–HLA DSA significantly reclassified patients at lower or higher risk for allograft loss with a category–free net reclassification improvement (NRI) of 0.494 (95% CI, 0.28 to 0.71; P<0.001).

Hierarchical Ranking of Anti–HLA DSA Characteristics on the Basis of Their Ability to Classify Patients According to their Risk of Allograft Loss

At the Time of Transplantation

In patients with anti-HLA DSA detected at the time of transplantation (n=110), we ranked day 0 iDSA characteristics on the basis of their ability to classify patients according to their risk of allograft loss by performing multivariate random survival forest analysis, which integrated all of the day 0 iDSA characteristics (iDSA HLA class, iDSA MFI level, iDSA C1q binding capacity, and iDSA IgG1–4 subclasses). The relative variable importance values were as follows: 1 for IgG3 positivity, 0.59 for C1q binding capacity, 0.16 for IgG4 positivity, 0.12 for IgG2 positivity, 0.11 for MFI, −0.02 for IgG1, and −0.03 for HLA class (Figure 2A).

Figure 2.

Hierarchical ranking of anti–HLA iDSA characteristics on the basis of their ability to classify patients according to their risk of allograft loss using random survival forest modeling. (A) At the time of transplantation (n=110). (B) Post-transplantation (n=186).

Post-Transplantation

The multivariate random survival forest analysis performed on patients with post–transplant anti–HLA DSA (n=186) (Figure 2B) showed the following relative variable importance values: 1 for IgG3 positivity, 0.42 for C1q binding, 0.16 for IgG1 positivity, 0.16 for HLA class, 0.06 for MFI, −0.04 for IgG4 positivity, −0.05 for IgG2 positivity, and −0.07 for preformed/de novo status.

Value of Anti–HLA DSA IgG3 Positivity and C1q Binding Capacity to Predict Allograft Loss at a Population Level

Because iDSA IgG3 positivity and C1q binding capacity were the most informative anti–HLA iDSA characteristics, we assessed their discrimination performance for the prediction of allograft loss in the overall study population (n=851).

At the Time of Transplantation

The performances of day 0 anti–HLA DSA, day 0 IgG3–positive iDSA, and day 0 C1q binding iDSA in predicting clinical and subclinical ABMR are shown in Table 4.

Table 4.

Performance of anti-HLA DSA, IgG3–positive anti–HLA iDSA, and C1q binding anti–HLA iDSA to predict clinical and subclinical ABMR in an unselected population of kidney transplant recipients (n=851)

Measures of Diagnostic Accuracy	Day 0 DSA, %	Day 0 IgG3 DSA, %	Day 0 C1q DSA, %	Post-Transplant DSA, %	Post–Transplant IgG3 DSA, %	Post–Transplant C1q DSA, %
Clinical ABMR
Sensitivity	55.4	33.8	32.3	100	58.5	52.3
Specificity	91.0	98.9	98.2	84.6	99.5	97.1
PPV	32.7	71.0	60.0	34.9	90.5	59.6
NPV	96.1	94.8	94.6	100	96.7	96.1
Subclinical ABMR
Sensitivity	49.3	4.5	11.9	100	4.5	29.9
Specificity	90.2	96.4	96.6	84.8	95.0	95.3
PPV	30.0	9.7	22.9	35.5	7.1	35.1
NPV	95.4	92.2	92.8	100	92.1	94.1

PPV, positive predictive value; NPV, negative predictive value.

Patients with day 0 IgG3–positive iDSA (n=31) had a decreased 5-year allograft survival (41.8%; 95% CI, 23.6 to 59.0) compared with patients without day 0 IgG3–positive iDSA (n=820; 90.8%; 95% CI, 88.5 to 92.7; P<0.001) (Supplemental Figure 1A). Patients with day 0 C1q binding iDSA (n=35) had a 5-year allograft survival of 51.0% (95% CI, 32.4% to 66.8%) compared with 90.7% (95% CI, 88.4 to 92.6) in patients without day 0 C1q binding iDSA (n=816) (Supplemental Figure 1B).

The addition of day 0 iDSA IgG3 positivity to the day 0 reference model yielded a c statistic of 0.72 (95% CI, 0.66 to 0.77; mean difference of 0.047; 95% CI, 0.045 to 0.048), whereas the addition of day 0 iDSA C1q binding capacity provided a c statistic of 0.70 (95% CI, 0.66 to 0.76; mean difference of 0.033; 95% CI, 0.032 to 0.035). The addition of both day 0 iDSA C1q binding capacity and IgG3 positivity to the day 0 reference model resulted in a c statistic of 0.74 (95% CI, 0.69 to 0.79).

Post-Transplantation

The performances of post–transplant anti–HLA DSA, post–transplant IgG3–positive iDSA, and post–transplant C1q binding iDSA in predicting clinical and subclinical ABMR are detailed in Table 4.

Patients with post–transplant IgG3–positive iDSA (n=42) showed a 5-year allograft survival of 30.0% (95% CI, 16.4 to 44.7) compared with 92.1% (95% CI, 89.9 to 93.9) in patients without post–transplant IgG3–positive iDSA (n=809) (Supplemental Figure 1C). Patients with post–transplant C1q binding iDSA (n=57) had a 5-year allograft survival of 45.8% (95% CI, 31.9 to 58.8) compared with 92.1% (95% CI, 89.9 to 93.9) in patients without post–transplant C1q binding iDSA (n=794) (Supplemental Figure 1D).

The addition of post–transplant iDSA IgG3 positivity and post–transplant iDSA C1q binding capacity to the post–Tx DSA model resulted in c statistics of 0.76 (95% CI, 0.72 to 0.82; mean difference of 0.046; 95% CI, 0.045 to 0.048) and 0.76 (95% CI, 0.72 to 0.82; mean difference of 0.045; 95% CI, 0.044 to 0.046), respectively. The addition of both post–transplant iDSA C1q binding capacity and IgG3 positivity to the post–Tx DSA model resulted in a c statistic of 0.81 (95% CI, 0.76 to 0.85).

Figure 3 depicts the predictive value for allograft loss of a strategy on the basis of systematic monitoring and precise characterization of anti-HLA DSA at the time of transplantation and after transplantation at the population level.

Figure 3.

Predictive value for allograft loss of a strategy on the basis of a systematic monitoring of anti-HLA DSAs and integration of anti–HLA DSA characteristics in an unselected population of kidney transplant recipients (n=851). Predictive value for allograft loss was assessed by Cox model Harrell c statistics in the overall study population (n=851). Day 0 anti–HLA DSA characteristics (IgG3 positivity and C1q binding) were added to the day 0 reference model, which was on the basis of a conventional strategy. Post–transplant anti–HLA DSA characteristics (IgG3 positivity and C1q binding) were added to the post–Tx DSA model. In the day 0 reference model and the post–Tx DSA model, anti-HLA DSAs were detected using the single–antigen Luminex technique. A c statistic of 0.5 indicated that the model is no better than chance at predicting membership in a group, and a value of one indicates that the model perfectly identifies those within a group and those not in a group. Percentile 95% CIs for c statistics were derived using 1000 bootstrap samples. The differences in c statistics were replicated 1000 times using bootstrap samples to derive 95% CIs.

Incremental Effect of Anti–HLA DSA IgG3 Positivity and C1q Binding Capacity on MFI Level for Stratifying the Individual Risk of Allograft Loss in Patients with Anti-HLA DSA

At the Time of Transplantation

In patients with anti-HLA DSA detected at the time of transplantation (n=110), the day 0 iDSA MFI level showed a c statistic of 0.75 (95% CI, 0.66 to 0.83). The addition of day 0 iDSA IgG3 positivity improved day 0 iDSA MFI discrimination performance, yielding a c statistic of 0.85 (95% CI, 0.79 to 0.92) with a mean difference of 0.105 (95% CI, 0.102 to 0.108) and an IDI of 0.233 (95% CI, 0.12 to 0.35; P<0.001) (Figure 4A). Day 0 iDSA IgG3 positivity adequately reclassified patients at lower or higher risk for allograft loss compared with iDSA MFI level alone, resulting in a category-free NRI of 1.304 (95% CI, 0.94 to 1.67; P<0.001). The addition of day 0 iDSA IgG3 positivity to day 0 iDSA MFI level reclassified 75 of 89 patients (84.3%) in the correct direction in the group without allograft loss by predicting a lower probability of allograft loss, whereas it adequately reclassified 17 of 21 patients (81.0%) in the group with allograft loss by predicting a greater probability of allograft loss (Figure 5A).

Figure 4.

Improvement in calculated risk of allograft loss by considering IgG3 and C1q binding anti–HLA DSA status in addition to anti–HLA DSA MFI level at (A) the time of transplantation and (B) post-transplantation. Improvement in calculated risk of allograft loss was assessed by the IDI. The IDI integrates the change in mean predicted probability of allograft loss in patients with allograft loss and those without allograft loss. The change in the mean predicted probability of allograft loss is adequate if it is positive for patients with allograft loss (increased calculated risk) and negative for those without allograft loss (decreased calculated risk). Tx, transplant.

Figure 5.

Individual additive value of IgG3 and C1q binding anti–HLA DSA status to MFI level for stratifying the risk of allograft loss at the time of transplantation ([A] IgG3 status and [B] C1q binding status) and post-transplantation ([C] IgG3 status and [D] C1q binding status). Additive value of IgG3 and C1q binding anti–HLA DSA status to MFI level was assessed by category-free NRI. The NRI integrates the direction of change in the probability of allograft loss for every individual. The change in individual calculated risk is in the correct direction if it is greater for patients with allograft loss and less for those without allograft loss. Blue lines in patients without allograft loss indicate that IgG3 and C1q binding anti–HLA iDSA status moved the individual predicted probability of allograft loss in the correct (downward) direction. Red lines in patients with allograft loss indicate a correct (upward) change in the predicted probability of allograft loss when adding IgG3 and C1q binding anti–HLA iDSA status to anti–HLA iDSA MFI level. Pts, patients; Tx, transplant.

The addition of day 0 iDSA C1q binding capacity to day 0 MFI level increased the c statistic to 0.79 (95% CI, 0.69 to 0.90; mean difference of 0.040; 95% CI, 0.038 to 0.042) and provided an IDI of 0.123 (95% CI, 0.06 to 0.19; P<0.001) (Figure 4A). Day 0 iDSA C1q binding capacity improved patient classification according to the patients’ risk of allograft loss with a category-free NRI of 0.929 (95% CI, 0.49 to 1.37; P<0.001). The addition of day 0 iDSA C1q binding capacity to day 0 iDSA MFI level reclassified 71 of 89 patients (79.8%) in the correct direction in the group without allograft loss, whereas it adequately reclassified 14 of 21 patients (66.7%) among the patients with allograft loss (Figure 5B).

The addition of both day 0 iDSA C1q binding capacity and IgG3 positivity to day 0 iDSA MFI level provided a c statistic of 0.87 (95% CI, 0.81 to 0.93).

Post-Transplantation

In patients with post–transplant anti–HLA DSA (n=186), iDSA MFI level had a c statistic of 0.73 (95% CI, 0.65 to 0.80). The addition of post–transplant iDSA IgG3 positivity increased the c statistic to 0.86 (95% CI, 0.80 to 0.91; mean difference of 0.130; 95% CI, 0.128 to 0.132). The IDI was 0.328 (95% CI, 0.24 to 0.42; P<0.001) (Figure 4B), and the category-free NRI was 1.326 (95% CI, 1.03 to 1.62; P<0.001). The addition of post–transplant iDSA IgG3 positivity to post–transplant iDSA MFI level reclassified 135 of 149 patients (90.6%) in the correct direction in patients without allograft loss, whereas it adequately reclassified 28 of 37 patients (75.7%) in patients with allograft loss (Figure 5C).

The addition of post–transplant iDSA C1q binding capacity to post–transplant iDSA MFI level provided a c statistic of 0.74 (95% CI, 0.65 to 0.83; mean difference of 0.016; 95% CI, 0.014 to 0.018). The IDI was 0.190 (95% CI, 0.12 to 0.26; P<0.001) (Figure 4B), and the category-free NRI was 0.948 (95% CI, 0.62 to 1.28; P<0.001). The addition of post–transplant iDSA C1q binding capacity to post–transplant iDSA MFI level reclassified 127 of 149 patients (85.2%) in the correct direction in patients without allograft loss, whereas it adequately reclassified 23 of 37 patients (62.2%) in patients with allograft loss (Figure 5D).

The addition of both post–transplant iDSA C1q binding capacity and IgG3 positivity to post–transplant iDSA MFI level provided a c statistic of 0.87 (95% CI, 0.82 to 0.92).

Sensitivity Analyses

The robustness of our results was confirmed by a sensitivity analysis performed after excluding patients with preformed anti–HLA DSA (n=110). In the population of patients without preformed anti-HLA (n=741), post–transplant anti–HLA DSA monitoring improved the day 0 reference model discrimination performance (Supplemental Table 3). The addition of post–transplant anti–HLA iDSA C1q binding capacity and IgG3 capacity further improved the model discrimination performance (Supplemental Table 3). In the patients with de novo anti–HLA DSA (n=105), anti–HLA iDSA C1q binding capacity and IgG3 positivity provided better discrimination of allograft loss when added to iDSA MFI level than MFI level alone (Supplemental Table 3).

Discussion

In this prospective study performed in an unselected population of 851 kidney transplant recipients, we showed that standardized monitoring of circulating anti–HLA DSAs within 2 years after transplantation improved the risk stratification for allograft loss. Extensive characterization of anti-HLA DSAs allowed us to show that IgG3 subclass positivity or complement binding capacity further improved pre- and post-transplant performance in predicting kidney allograft loss beyond the conventional approach on the basis of the detection of circulating anti–HLA DSAs and the assessment of their strength using Luminex technology. We showed that a precise characterization of anti-HLA DSAs, including IgG3 status or C1q binding status, improved the evaluation of individual risk for allograft loss in >60% of patients.

Our study showed that a significant proportion of patients (100 of 186; 53.8%) did not show allograft dysfunction at the time of post–transplant anti–HLA DSA detection. In these clinically stable patients, concurrent kidney allograft biopsies revealed acute/active or chronic/active ABMR in 67 (67.0%) patients, emphasizing the importance of performing allograft biopsy at the time of anti–HLA DSA detection in uncovering subclinical ABMR disease.³² Systematic monitoring of anti-HLA DSAs might allow for the early diagnosis of ABMR disease and subsequent specific treatment and adjustment of immunosuppressive therapy.

This study extended the results of previous work showing relationships between circulating anti–HLA DSA strength and allograft lesion intensity and allograft survival, with higher levels of circulating anti–HLA DSAs being associated with increased microvascular inflammation, increased C4d deposition in the peritubular capillaries of the allograft,^33–35 and decreased allograft survival.² More recently, other properties of anti-HLA DSAs have been associated with kidney allograft loss, including anti–HLA DSA complement binding capacity and IgG subclass composition.^{14,15,18,19,25–27} However, the predictive value of anti–HLA DSA characteristics for kidney transplant outcomes assessed at the time of transplantation and after transplantation had not been previously investigated accurately.^30,36 This study addressed, for the first time, the dynamic and incremental prediction of kidney allograft loss using a prospective systematic monitoring and characterization of anti-HLA DSA (MFI, C1q binding capacity, and IgG subclasses) together with its potential for individual risk reclassification.

Using dedicated analyses, we evaluated the added predictive ability of the most informative anti–HLA DSA characteristics (IgG3 and C1q binding status) for the reclassification of individual risk of allograft loss. The increase in c statistic showed that the addition of anti–HLA DSA IgG3 and C1q binding status to anti–HLA DSA MFI level improved the concordance between predicted and observed kidney allograft survival. The IDI showed significant improvement in the magnitude of the change in the predicted risk of allograft loss when adding anti–HLA DSA IgG3 and C1q binding status to anti–HLA DSA MFI level, whereas the NRI determined a significant change in the adequate direction of the individual predicted risk of allograft loss.

The risk-stratified approach greatly increases our ability to personalize the clinical management of patients with kidney transplants.³¹ Our results suggest that the risk of ABMR and allograft loss might be significantly reduced by avoiding HLA-incompatible transplant across preformed C1q binding and/or IgG3–positive anti–HLA DSA. In hypersensitized patients with an insufficient flow of donors, specific pretransplant conditioning should be considered to eliminate C1q binding and/or IgG3–positive anti–HLA DSA before accepting a transplant. In the post-transplant setting, the systematic monitoring and the characterization of anti-HLA antibodies provide noninvasive tools to identify patients who are at high risk of ABMR and allograft loss. Furthermore, risk assessment on the basis of anti–HLA DSA properties could provide a basis for more personalized pathogenesis–driven therapies. Patients with anti-HLA DSA showing complement binding ability reflected by C1q or IgG3 positivity might benefit from more specific therapeutic protocols using complement-targeting agents.³⁷ More generally, risk stratification should be greatly integrated in randomized, controlled trials to apply averaged results of clinical trials to individual patients,³⁸ because their aggregated results can be misleading when applied to individual patients.³¹

The principal limitation of this study pertained to the fact that we were not able to validate our findings in an independent population given the uniqueness of our prospective, highly phenotyped cohort integrating systematic immunologic monitoring, protocol biopsies, and extensive assessment of anti-HLA DSA. Furthermore, specific economic studies are needed to evaluate the cost efficiency of systematic anti–HLA DSA monitoring policies in kidney transplantation before translation in clinical routine.

In conclusion, this prospective study, performed in a large cohort of kidney transplant recipients with systematic anti–HLA DSA screening and allograft biopsies, showed that post-transplant monitoring of circulating anti–HLA DSA using the single–antigen flow bead technique improved the individual risk stratification for allograft loss. We also showed that the addition of IgG3 or C1q binding anti–HLA DSA status to the conventional approach on the basis of anti–HLA DSA strength improved the performance in assessing the individual risk for allograft loss in >60% of patients.

Concise Methods

Study Design

We enrolled all consecutive patients who underwent kidney transplantation at Saint Louis Hospital (n=429) and Necker Hospital (n=477) between January 1, 2008 and December 31, 2010 (n=906). Patients were followed until January 1, 2016. All of the transplants were ABO blood group compatible and performed with negative standard National Institutes of Health and anti–human globulin T and B cell cytotoxicity crossmatches. Patients transplanted after desensitization protocols (n=21) and those enrolled in clinical trials (n=34) were excluded. All of the included patients (n=851) were screened for the presence of circulating anti–HLA DSA (1) at the time of transplantation, (2) systematically at 1 and 2 years after transplantation, and (3) at the time of a clinical event occurring within the first 2 years after transplantation (Figure 1).

All study patients identified with circulating anti–HLA DSA were tested for anti–HLA DSA characteristics, namely DSA specificity, DSA HLA class, DSA MFI level, DSA C1q binding capacity, and DSA IgG1–4 subclass, at two time points: at the time of transplantation and at the time of allograft biopsy (as detailed in Figure 1).

The study was approved by the Institutional Review Boards of Saint Louis Hospital and Necker Hospital.

One Lambda, Inc. (Canoga Park, CA) donated reagents but was not otherwise involved in either the conduct of the study or the preparation of the manuscript.

Clinical Data

The clinical data regarding donors and recipients were extracted from a prospective database: Données Informatiques Validées en Transplantation (DIVAT; www.divat.fr). Coding was used to ensure strict donor and recipient anonymity. The data are computerized in real time and at each transplant anniversary, and they are submitted for an annual audit. Each patient in this study provided written informed consent to be included in the DIVAT database network. This database is approved by the National French Commission for Bioinformatics Data and Patient Liberty (Commission Nationale de l'Informatique et des Libertés registration no. 1016618; validated June 8, 2004).

Clinical events were defined by the following: (1) an increase in serum creatinine exceeding 15% of the baseline value within a period of 21 days without ultrasound abnormalities, (2) proteinuria exceeding 0.5 g/g, and (3) an increase in MFI level exceeding 50% compared with the day 0 level in patients with preformed anti–HLA DSA.

Renal function was assessed by the eGFR with the abbreviated Modification of Diet in Renal Disease formula³⁹ at the time of post–transplant anti–HLA DSA detection.

The immunosuppression protocols and treatment of allograft rejections after transplantation were similar between the centers. The protocols and treatments are described in Supplemental Material.

Detection and Characterization of Anti-HLA DSAs

Screening for Anti-HLA DSAs in the Study Population

All of the kidney transplant recipients (n=851) were tested for circulating anti–HLA-A, -B, -Cw, -DR, -DQ, and -DP DSAs in serum samples obtained at the time of transplantation, systematically at 1 and 2 years after transplantation, and at the time of a clinical event occurring in the first 2 years after transplantation. All of the serum samples were treated with EDTA; a 0.1 M solution of disodium EDTA at pH 7.4 was diluted 1:10 in the serum and incubated for 10 minutes before testing. Single–antigen flow bead assays were used (One Lambda, Inc.) on a Luminex platform. All beads showing a normalized MFI >1000 were considered positive. The highest MFI value toward a donor-specific allele was considered to be the iDSA.

HLA typing of all of the kidney transplant donors and recipients was performed by molecular biology (Innolipa HLA Typing Kit; Innogenetics, Gent, Belgium).

Antibody Characterization in the Patients with Positive Screening for Anti-HLA DSAs

Serum samples from patients with circulating anti–HLA DSA at the time of transplantation (n=110) and the time of the first post–transplant circulating anti–HLA DSA detection (n=186) were analyzed in a blinded fashion at the University of Pittsburgh for the presence of C1q binding anti–HLA DSA and the presence of IgG1–4 subclasses.

The presence of C1q binding anti–HLA DSAs was assessed using single–antigen flow bead assays according to the manufacturer’s protocol (C1q Screen; One Lambda, Inc.).

The IgG subclass assay was performed as reported previously²⁵ using a modified standard single–antigen assay and replacing the phycoerythrin–conjugated antipan–human IgG reporter antibody with mAbs specific for IgG1–4 subclasses (IgG1 clone HP6001, IgG2 clone 31–7-4, IgG3 clone HP6050, and IgG4 clone HP6025; Southern Biotech).

The specificity of the iDSA on the basis of pan-IgG reactivity was applied to the other tests, including C1q binding and IgG subtype analysis.

For each patient, we evaluated the number, HLA class, and MFI of all of the detected anti-HLA DSAs, and for iDSAs, we also considered the C1q binding capacity and the IgG1–4 subclasses.

Post–transplant anti–HLA DSAs were considered preformed when they were detectable at the time of transplantation and persisted after transplantation. If anti-HLA DSAs were absent at the time of transplantation as determined by solid-phase assay and became detectable post-transplant, they were considered de novo DSAs.

Kidney Allograft Histology

Kidney allograft biopsies were performed at the time of the first post–transplant anti-HLA DSA detection in patients with de novo anti–HLA DSA (clinical event or 1 or 2 years after transplantation). In patients with preformed anti–HLA DSA, allograft biopsies were performed at the time of a clinical event (in patients with allograft dysfunction, proteinuria, or an MFI increase exceeding 50%) or 1 year after transplantation in stable-state patients with persistent anti–HLA DSA. Renal tissue was fixed in acetic formol absolute alcohol fixative and stained with Masson trichrome and periodic acid–Schiff. All of the graft biopsies were scored and graded from zero to three according to the updated Banff criteria^40–43 for the following histologic factors: glomerulitis, tubulitis, mononuclear cell interstitial inflammation, intimal arteritis, peritubular capillaritis, allograft glomerulopathy, interstitial fibrosis/tubular atrophy, arteriolar hyaline thickening, and vascular fibrous intimal thickening. C4d staining was performed by immunochemical analysis on paraffin sections using polyclonal human anti–C4d antibodies (Biomedica Gruppe, Vienna, Austria).

All of the graft biopsies were scored and graded by experienced pathologists (C.G. and J.-P.D.v.H.) who were unaware of the patients’ clinical and immunologic statuses.

ABMR was classified according to the last update of the Banff classification.⁴³ Subclinical ABMR was defined by stable renal function and the Banff criteria for acute/active or chronic/active ABMR. Stable renal function was defined as the variability in serum creatinine not exceeding 15% of the baseline value within a period of 21 days before the biopsy.³²

Statistical Analyses

The mean±SD values and frequencies are provided for the description of the continuous and categorical variables, respectively, unless otherwise stated. The means and proportions were compared using the t test and the chi-squared test, respectively (or Mann–Whitney U test and Fisher exact test if appropriate, respectively).

First, we built the day 0 reference model by assessing the determinants of time to kidney allograft failure at the time of transplantation among traditional risk factors, including recipient, donor, transplant characteristics, and the presence of anti-HLA DSA detected by single-antigen Luminex. Kidney allograft survival was calculated from the date of transplantation to the date of allograft loss. In patients who died with a functioning graft, graft survival was censored at the time of death. Graft survival analyses were performed for a maximum follow-up period of 5 years from the time of transplantation. Cox proportional hazard models were used to estimate the HRs and 95% CIs for kidney allograft loss. We first performed univariate analysis. A P value threshold of 0.20 for entering variables into the multivariate model was used. Significant risk factors were then entered into a single multivariate model using backward stepwise elimination to define the day 0 reference model. Model calibration and goodness of fit were assessed by visual examination of a calibration plot.

Second, we assessed the change in the discrimination capacity of the day 0 reference model by adding the detection of post–transplant anti–HLA DSA by single-antigen Luminex (post–Tx DSA model). Harrell c statistic was estimated for the day 0 reference model and the post–Tx DSA model; c-statistic estimations were repeated 1000 times using bootstrap samples to derive 95% CIs and assess the difference in the c statistic between the models with its 95% CI. We used category-free NRI and IDI to assess the incremental effect of post–transplant anti–HLA DSA detection on the day 0 reference model to predict allograft loss.^36,44

Third, we hierarchically ranked the anti–HLA iDSA characteristics at the time of transplantation and after transplantation (iDSA HLA class, iDSA MFI, iDSA C1q binding capacity, and iDSA IgG1–4 subclasses) according to their ability to classify subjects who would lose their graft from those who would not by performing multivariate random survival forest modeling. Five thousand trees were generated using bootstrapping by sampling with replacement at the root node before growing trees. The following parameters were applied: the minimum number of unique patients in a terminal node was set at three, the split rule was log-rank splitting, three variables (square root of the number of variables rounded up) were randomly selected as candidates for each node split, and the maximum number of split points randomly chosen among the possible split points for each variable was set at one. Variable importance was calculated using Breiman–Cutler permutation variable importance.⁴⁵

Fourth, we assessed the change in the discrimination capacities of the day 0 reference model and the post–Tx DSA model by adding the top anti–HLA iDSA characteristics identified in survival forest modeling at the time of transplantation and after transplantation. Harrell c statistic was estimated for the day 0 reference model, the post–Tx DSA model, and the additive Cox models; c-statistic estimations were repeated 1000 times with the use of bootstrap samples to derive 95% CIs and assess the differences in the c statistic between models with their respective 95% CIs.

Fifth, we used category-free NRI and IDI to assess the reclassification of the risk of allograft loss when adding anti–HLA iDSA IgG3 status and C1q binding status to iDSA MFI level at the time of transplantation and post-transplantation.^36,44 Model calibration and goodness of fit were assessed by visual examination of a calibration plot.

The values of P<0.05 were considered statistically significant, and all of the tests were two sided. All analyses were performed using STATA software, version 12 (StataCorp., College Station, TX) and R software, version 2.15.2 (R Development Core Team).

Details on the interpretation of important statistical methods are provided in Supplemental Material.

Disclosures

None.