The present study was approved by the Stanford University Institutional Review Board (49,338). The clinical and research activities reported herein are consistent with the principles of the Declarations of Helsinki and Istanbul.

Pediatric patients who underwent kidney transplantation at Stanford University’s LPCH from January 1st, 2010 to March 1st, 2018 were eligible for the study. Patients with less than 12 months of follow-up, multiorgan transplants, or inconclusive DSA data were excluded. If a patient received more than one transplant during the study period, only data on the first allograft were used.

Data were retrospectively extracted from the electronic medical record system, UNOS^® and the Stanford Histocompatibility, Immunogenetics, and Disease Profiling Laboratory electronic database at LPCH. Information on patient characteristics, such as cause of end-stage kidney disease, donor information, HLA matching, and age at transplant, immunosuppressive treatment and allograft function were collected. Induction of immunosuppression protocol at LPCH included a rabbit antithymocyte globulin. Maintenance immunosuppression consisted of tacrolimus and mycophenolate mofetil (MMF) for all patients, with or without prednisone, based on immunologic risk. Serum creatinine and urine protein/creatinine ratio were measured at least every 3 months.

De novo DSA formation was defined as donor-specific anti-HLA antibodies that were initially identified after the kidney transplant. During the study period, all patients in our kidney transplant program were tested for both standard dnDSA and C1q-dnDSA at 0 (time of transplant), 1, 2, 3, 6, and 12 months following transplant surgery, at least annually thereafter, and as clinically indicated (e.g., in the case of allograft dysfunction). Information was collected on all dnDSA, including A, B, C, DR, DQ and DP specificities. At the Stanford HLA laboratory, commercially available Single Antigen Bead (SAB) assay kits (LAB Screen; One Lambda, Inc., Canoga Park, CA, United States) were used for the detection of antibodies. Standard dnDSA were defined as HLA IgG antibodies identified by the solid phase assay on a Luminex platform (23).

The presence of C1q-dnDSA was determined using a SAB assay according to the manufacturer’s protocol (C1qScreen™, One Lambda Inc.) on a Luminex platform. In brief, patients’ sera were mixed with polystyrene beads, each uniquely distinguishable by subtle differences in fluorochromes and each coated with a different purified, single‐cloned HLA class I or class II antigen. Data were analyzed using the HLA Fusion™ software (One Lambda Inc.), and interpretations were made using normalized (baseline) mean fluorescence intensity (MFI) values. Cutoffs for positive reactions were >1,000 MFI. DSA were considered to have disappeared if such antibodies were <1,000 MFI at the end of follow-up or at the time of graft failure. Persistence of DSA was defined as any DSA with MFI >1,000 at the end of follow-up or at the time of graft failure, even if they had transiently become negative at any point. Immunodominant DSA (iDSA) was defined as the highest MFI value of standard dnDSA. An important methodological issue is how to analyze and report the various patterns of antibody response in a heterogeneous patient group. Some patients may generate an antibody response to 1 or 2 HLA antigens, but with a very high MFI. Other patients might generate antibodies to multiple, even dozens of HLA antigens, but with intermediate or lower MFI. As an approach to this issue we used Jordan’s previously published Relative Intensity Score (RIS) (24–26), in addition to MFI levels. To calculate RIS, we scored combined MFI of all DSA in the following manner: Each DSA with MFI <1,000 received 0 points; MFI 1000–5,000 (weak intensity) received 2 points; MFI 5,000–10,000 (moderate intensity) received 5 points; and MFI >10,000 (strong intensity) received 10 points. The points were summed to form the RIS score.

During the period of the study, protocol biopsies were performed at 6, 12, and 24 months after kidney transplantation and as clinically indicated (i.e., if graft dysfunction and/or dnDSA appearance). To ensure consistency, transplant biopsy specimens were gathered and scored for analysis by an expert in kidney transplant pathology according to the consensus rules of the most recent international Banff Classification criteria (27). Baseline immunosuppression and treatment protocol for ABMR after detection of C1q-dnDSA, with or without biopsy, remained unchanged throughout the study period. With the initiation of standardized DSA monitoring, including complement-biding capacity, we implemented an augmented immunosuppressive treatment protocol for all patients with C1q-dnDSA MFI >1,000 shown in Table 1. This protocol was unchanged throughout the study period. Patients who formed standard dnDSA only were not treated.

Donor-specific antibody status was considered as the exposure. C1q-dnDSA was a time-fixed variable of C1q status, categorized as negative, eliminated or persistent. If C1q-dnDSA disappeared after detection at some point and the MFI was <1,000 for the remainder of the study, the C1q-dnDSA were considered “eliminated.” In those who continued to have C1q-dnDSA throughout the study or at the time of allograft failure, the C1q-dnDSA were categorized as “persistent.” Standard dnDSA were either negative, present without ever binding complement or detected together with C1q-dnDSA. The persistence and elimination of standard dnDSA was followed in patients with C1q-dnDSA and defined in the same manner as above.

Decreased immunosuppressive therapy as a risk factor for dnDSA formation was defined as purposeful reduction of immunosuppression in response to side effects, malignancy or infectious concerns. Nonadherence to medical care was defined when documented in the medical chart, when patients had undetectable tacrolimus levels, missed clinic visits, missed blood draws, or by patient report. Allograft failure was defined as return to dialysis or preemptive re-transplantation. Decreased GFR was defined as estimated GFR <60 ml/min/1.73 m² that persisted over at least 3 months. The creatinine-based “Bedside Schwartz” equation (2009) and/or Chronic Kidney Disease Epidemiology Collaboration equation were used to calculate GFR. Proteinuria cutoff was set at urine protein/creatinine ratio of 0.5 mg/mg that persisted over 3 months. Patients who did not have adverse graft outcome were censored at their last follow up. No patient died before the endpoints of interest were reached.

Baseline demographic and clinical characteristics of the cohort were summarized descriptively using means (range), medians (interquartile range, IQR), and counts (percentages) as appropriate. Median time-to-allograft failure was reported using Kaplan-Meier estimates. The Kruskal-Wallis, Mann-Whitney, Chi-squared, and Fisher’s exact tests were used to compare clinical characteristics of patients with persistent C1q-dnDSA to those of patients with eliminated C1q-dnDSA as appropriate based on underlying statistical assumptions. A conditional inference forest analysis was employed to assess the hierarchy of the characteristics of de novo anti-HLA DSA based on their ability to predict adverse graft outcome, defined by decreased GFR or allograft failure (28). Included in the model were iDSA MFI, RIS of standard DSA, C1q-binding status, and standard dnDSA status. The variables included in the random forest approach were set to zero for patients who did not form anti-HLA dnDSA. The conditional forest was fit to 1,000 trees. Given the collinearity among features, conditional variable importance measures were computed in order to quantify the contribution of each variable, using the integrated Brier score as a risk measure (29). Out-of-bag model performance statistics were expressed for the binary outcome of adverse graft event.

Outcomes of graft failure and proteinuria were analyzed based on time-varying C1q-dnDSA-binding status. An unadjusted Cox proportional hazards regression model was fit to time-to-allograft failure as a function of time-varying C1q binding. The proportional hazards assumption was assessed for each model and found to hold in all cases (30). Hazard ratios (HR), 95% confidence intervals (CI), and model concordance (using the C-index) are reported (31). Covariates of interest included sex, historic peak panel-reactive antibodies (pPRA), age at transplant, type of donor and HLA match. A p-value <0.05 was considered statistically significant. All analyses were conducted using R version 3.5.2.

Out of 288 patients who underwent kidney transplantation at LPCH, 233 met the inclusion criteria. One patient who experienced graft failure before 12 months posttransplant due to BK nephropathy was included. Mean follow-up time was 45 (range, 9–108) months. The median age at transplant was 12.3 years, and the majority of the transplant recipients were of Hispanic/Latino origin (39%) and male sex (53%) (Table 2). A little over half (54%) had an HLA match of 0 or 1 out of 6, with only 6% having an HLA match >3/6. The median pPRA was 7%, and 76% of donors were deceased. The most common cause of end-stage kidney disease was renal aplasia/hypoplasia/dysplasia (22%).

Among the 233 study subjects, 118 formed standard dnDSA and of those, 52 also had complement-binding activity triggering interventions at initial detection of C1q-dnDSA. A flowchart illustrating the study cohort is provided in Figure 1. C1q-dnDSA-directed treatment was individualized based on MFI strength, clinical factors, and biopsy results if available, according to the augmented immunosuppressive therapy protocol. The iDSA were of HLA class 1 in 19 (16.1%) patients and HLA class 2 in 99 (83.8%) patients. The median peak MFI of iDSA in all patients forming anti-HLA dnDSA was 4,072 (IQR, 11,304). The iDSA were of DQ or DR specificity in 94 (79.6%) patients. Sixty-six (28.3%) patients formed standard dnDSA without C1q binding with a median peak iDSA MFI of 1755 (IQR, 2052), and 52 (22.3%) patients formed C1q-dnDSA with the median peak iDSA MFI of 13,665 (IQR, 11,317). The median peak RIS of standard dnDSA, calculated for patients who also had C1q-DSA, was 20 (IQR, 22).

All 233 patients were included in the random forest analysis. The most important characteristic for an adverse graft outcome (n = 46) was the persistence of C1q binding (Figure 2). Seventeen out of 21 (80.9%) patients with C1q persistence had an adverse graft outcome, 14 of whom lost their graft. In patients without DSA formation, 16 (13.9%) had an adverse graft outcome, including two who lost their graft. Eight (12.1%) of the patients with standard dnDSA without complement binding and 5 (16.1%) of those with standard dnDSA and disappearance of C1q after treatment experienced an adverse graft outcome, although none lost their graft. Adverse graft outcomes were significantly more frequent among patients with C1q persistence compared to others (p <0.0001), but were not significantly different between non-DSA formers, standard dnDSA formers only and C1q-dnDSA eliminated groups (p = 0.86). The random forest model correctly classified only 17 of the 46 (37%) patients as having an adverse graft outcome, while 184 of 187 (98.3%) were correctly classified as not having an event. Of 14 patients who lost their graft due to immunological reasons, 13 had HLA class 2 dnDSA. Of those with dnDSA and graft failure12 of 14 had DQ- or DR-specific iDSA. Maximum iDSA MFI was more useful to predict graft outcome than maximum RIS. Peak iDSA MFI was correlated with adverse outcome. Looking at the peak iDSA MFI among the 118 patients who formed standard dnDSA, patients with adverse graft outcome had significantly higher peak of 10,861 (IQR, 12,394) compared with 2,987 (IQR, 6,891) in those without adverse graft outcomes (p = 0.01). If we only analyze the C1q formers, there was no difference between peak iDSA MFI levels in patients with or without graft failure, 13,019 (IQR, 13,665) versus 13,569 (IQR, 7,421), respectively (Figure 3).

The 52 patients who formed C1q-dnDSA had a median time of 3.8 years (IQR, 1.3) until their first detection. C1q-dnDSA disappeared in 31 (59.6%) patients and persisted in 21 (40.4%). The time to detection was not different between patients with persistent C1q-dnDSA and those who eliminated the antibodies. Forty-four (84.6%) patients had a kidney allograft biopsy carried out at the time of detection, of those 33 (75%) had ABMR according to histological findings. Eight patients were not biopsied, of these four were treated according to protocol without a biopsy, two had contraindications to biopsy, 1 patient refused to have the biopsy performed, and 1 patient started dialysis shortly after detection of C1q. The clinical characteristics and histological findings of patients who formed C1q-dnDSA are shown in Table 3. The C1q persistent was more likely to have histological signs of ABMR on biopsy at the time of detection, including positive C4d and interstitial inflammation, compared with the C1q disappearance group. They were also older at kidney transplantation and more likely to be nonadherent to immunosuppressive treatment compared with patients who eliminated C1q. The latter group was younger at transplantation and at first detection of C1q; they were also more likely to have purposeful immunosuppressive therapy reduction as a risk factor for C1q formation. The evolution of DSA over time in the cohort is shown in Figure 4. Standard dnDSA RIS and C1q-dnDSA RIS were moderately correlated (Pearson correlation coefficient of 0.61). C1q-binding status varied and sometimes reappeared after a period of being undetectable. In the group where C1q disappeared, five patients had at least one C1q reappearance, whereas this was observed for nine patients in the group where C1q persisted. Two patients with C1q-dnDSA persistence were treated with plasmapheresis. All patients with persistent C1q had documented medication non-compliance. One patient with persistent C1q had both infections and history of non-compliance at the time of C1q detection. Two patients in whom C1q disappeared did not have an identifiable risk for C1q binding. Comparing the evolution of C1q-dnDSA and standard dnDSA, the latter stayed elevated longer and did not respond to enhanced immunosuppressive treatment in the same way as C1q-dnDSA. Patients with purposeful reduction of immunosuppression appeared to have a steeper decrease of C1q-dnDSA. In the patients with C1q-dnDSA, persistence of standard dnDSA was not strongly correlated with graft failure, unlike C1q persistence. Out of 31 patients who eliminated C1q, 15 (48.4%) had persistent standard-dnDSA compared with all 21 patients with persistent C1q-dnDSA. Among patients with persistent C1q-dnDSA, older age at transplant (HR, 3.69; 95% CI, 1.58–8.63) was significantly associated with graft failure. Younger age at first detection of C1q decreased the risk of graft failure (HR, 0.33; 95% CI, 0.15–0.72). Severe complications after treatment of dnDSA, such as malignancies, infections requiring admission to the intensive care unit, or hypogammaglobulinemia requiring intravenous immunoglobulin (IVIG), were not observed.

In the entire cohort, graft failure occurred in 16 (6.9%) patients after a median of 45 months (IQR, 28). Fourteen of these patients experienced allograft rejection and two had other potential causes of graft failure; one had BK nephropathy and the other cancer chemotherapy nephrotoxicity (Figure 5). All patients whose grafts failed due to immunological causes had persistent C1q-dnDSA in spite of augmented immunosuppressive treatment and lost their graft at a median of 51 months (IQR, 25). Persistent C1q-dnDSA were significantly associated with higher risk of graft failure (adjusted HR, 45.46; 95% CI, 11.7 to 177.4). In both the unadjusted and adjusted analysis, elimination of C1q-dnDSA was significantly associated with lower risk of graft failure (adjusted HR, 0.02; 95% CI, 0.01–0.09). Among patients with persistent C1q-dnDSA, older age at transplant (HR, 3.69; 95% CI, 1.58–8.63) and younger age at first detection of C1q (HR, 0.33; 95% CI, 0.15–0.72) were significantly associated with graft failure. No other demographic or clinical features were associated with graft failure in both groups. There was no correlation between peak iDSA and graft outcome or C1q persistence. The frequency of adverse graft outcome was not significantly different in patients with no dnDSA formation, standard-dnDSA only and eliminated C1q-dnDSA. The risk of proteinuria was highest in patients with persistent C1q-dnDSA, intermediate in those with transient C1q, and lowest in those who never developed C1q-dnDSA (Figure 6).