Molecular criteria for pulmonary antibody-mediated rejection are associated with an increased risk of allograft failure

Abstract

Background:

Current International Society for Heart and Lung Transplantation (ISHLT) criteria for pulmonary antibody-mediated rejection (AMR) is predicated on a constellation of clinical, laboratory and histopathological parameters, including the presence of donor-specific antibodies (DSA). However, molecular evidence of allograft injury is not considered. The aim of this study was to investigate if allograft injury on the molecular level, as measured by donor-derived cell-free DNA (dd-cfDNA), identifies DSA positive patients experiencing a form of AMR associated with increased risk of chronic lung allograft dysfunction (CLAD) or death.

Methods:

This multicenter, observational analysis included adult lung transplant recipients from two prospective cohort studies. Serial plasma samples were collected for dd-cfDNA measurement by shotgun sequencing. Molecular AMR was defined as the presence of DSA and dd-cfDNA level >1% occurring > 30 days post-transplant. Clinical AMR was defined using ISHLT criteria. Time-dependent multivariable Cox regression models were used to determine the association of Clinical AMR or Molecular AMR with the composite outcome of CLAD or death.

Results:

The final analysis included 209 subjects. Sixty-one subjects met criteria for molecular AMR. Molecular AMR captured 42/46 (91%) of patients who experienced Clinical AMR. Molecular AMR was associated with an increased risk of CLAD or death (HR 2.00, 95% CI: 1.18 – 3.38, p = 0.010). The results remained consistent analyzing Molecular AMR subjects without concomitant ISHLT Clinical AMR, acute rejection, or infection (HR 2.45, 95% CI: 1.01 – 5.94, p = 0.047).

Conclusions:

Molecular AMR identifies a population of lung transplant recipients potentially experiencing antibody-mediated rejection not captured by current ISHLT criteria.

Introduction

Despite advances in recognition, diagnostic criteria and treatment, pulmonary antibody-mediated rejection (AMR) in lung transplant recipients remains associated with the development of chronic lung allograft dysfunction (CLAD) and death(1). The current International Society for Heart and Lung Transplantation (ISHLT) criteria for pulmonary AMR is predicated on a constellation of clinical, laboratory and histopathological parameters, including the presence of donor-specific antibodies (DSA)(2). However, this definition may pose certain challenges. For instance, it relies on the coordinated performance of several diagnostic modalities within a short time frame, thereby posing logistical obstacles for expedient diagnosis. Further, the diagnosis of clinical AMR requires the presence of allograft dysfunction in the form of pulmonary function decline or hypoxemia, in which case AMR may be captured too late in the process to effectively treat the underlying condition. Histopathology and complement fixation staining are further limited by low sensitivity and/or high interrater variability. Alternative approaches to the diagnosis of AMR and the assessment of underlying allograft injury in the setting of humoral rejection may address these limitations.

Prior studies have demonstrated that donor-derived cell-free DNA (dd-cfDNA) increases in the setting of pulmonary allograft injury(3–8). Further, levels of dd-cfDNA may rise several weeks-to-months prior to the clinical manifestations of acute rejection, particularly AMR(3). This may allow for a more sensitive and timely approach to the diagnosis of AMR. The most recent ISHLT consensus statement on AMR emphasizes assessing the “need for molecular diagnostic criteria for pulmonary AMR” as a research priority(2). Therefore, in this study, we hypothesize that the combination of a molecular assessment of allograft injury (using dd-cfDNA) with DSA – to define Molecular AMR – will identify a population of patients at increased risk of CLAD or death not currently captured by current ISHLT criteria.

Methods

This is a multicenter, observational analysis that included lung transplant recipients > 18 years of age enrolled in two prospective cohort studies. Methodological details of these studies have been described previously(9, 10). The first study, Genome Research Alliance for Transplantation (GRAfT) (NCT0243070), began enrollment in 2015 and is ongoing at three centers (the Johns Hopkins Hospital, Inova Fairfax Hospital and University of Maryland Medical Center). We included all GRAfT patients enrolled prior to July 1^st, 2019 that underwent formal adjudication for AMR by the GRAfT AMR Adjudication Committee. The second study, Genome Transplant Dynamics (GTD) (NCT01985412), enrolled patients at Stanford University Hospital between December 1, 2010, and December 31, 2012, with follow up until May 1, 2019. Each study was designed to investigate the ability of dd-cfDNA to detect episodes of acute lung allograft dysfunction. Clinical data was collected during routine post-transplant monitoring with clinic visits, pulmonary function testing (PFT), surveillance bronchoscopy with bronchoalveolar lavage (BAL) and transbronchial biopsy (TBBx), and donor specific antibody testing (DSA). Patients underwent serial plasma sampling for dd-cfDNA on post-transplant days 1, 3, 7, 14, 21 and at the time of all surveillance and for-cause bronchoscopies. Patients that died/censored < 90 days post-transplant, had insufficient PFT data to adjudicate for CLAD or Clinical AMR, had no DSA or dd-cfDNA data > 30 days post-transplant were excluded from the study. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement guidelines for reporting observational studies were followed. This study was approved by the Institutional Review Board at each participating center and the National Heart, Lung, and Blood Institute.

DSA Testing and Measurement

The transplant centers performed surveillance DSA testing prior to transplantation and on post-transplant day 7, 14 and along with surveillance bronchoscopy on month 1, 3, 6, 9, 12, 18 and 24. Additional DSA testing was performed for clinical signs or symptoms of allograft dysfunction. Each center utilized single antigen bead testing using the LABScreen^® Single Antigen Bead assay (One Lambda, Canoga Park, CA) and designated DSA as either positive or negative. A positive test was defined as a mean fluorescence intensity (MFI) ≥ 1000 on one occasion or an MFI between 500–1000 on two serial occasions.

Definition of Molecular AMR and Clinical endpoints

The primary endpoint of this study was a composite of CLAD or death, whichever occurred first. CLAD was defined according to International Society for Heart and Lung Transplantation (ISHLT) criteria as a sustained decrease of ≥ 20% forced expiratory volume in one second (FEV₁) from baseline value at least 3 months post-transplant and taken at least 3 weeks apart(11). Acute rejection was categorized as acute cellular rejection (ACR) graded by experienced pathologists at each center according to ISHLT consensus guidelines(12) or clinical AMR adjudicated by a multidisciplinary adjudication committee. The committee was composed of two transplant pulmonologists, one internist, one transplant immunogeneticist, one transplant pharmacist and two pathologists. AMR was defined according to ISHLT guidelines for possible, probable, and definite clinical AMR(2). ISHLT Clinical AMR was defined as the presence of allograft dysfunction (acute decline in FEV1 ≥ 10%) and positive DSA plus one of the following: positive histopathology, positive c4d staining or absence of an alternative diagnosis(2).

Molecular AMR was defined as the presence of DSA (either pre-formed or de-novo) with a concomitant dd-cfDNA >1%. The threshold of 1% was selected based on prior studies demonstrating that levels >1% in the setting of acute lung allograft dysfunction predict poor long-term outcomes(13, 14). The time of Molecular AMR was defined as the time at which the dd-cfDNA first increased >1% in a DSA positive patient.

Measurement of dd-cfDNA

Plasma dd-cfDNA was measured using an automated shotgun sequencing method(15). First, donor and recipient pre-transplant genomic DNA was isolated and genotyped. Next, the donor and recipient genomes were compared to identify single-nucleotide polymorphisms (SNPs). Following transplantation, recipient plasma cell free DNA (cfDNA) was isolated to generate a DNA library for paired-end shotgun sequencing. The cfDNA sequence reads were next interrogated for donor and recipient SNPs and dd-cfDNA was calculated as the percentage of donor SNPs to total (recipient plus donor) SNPs. The value of dd-cfDNA in single lung transplant recipients was adjusted to account for differences in allograft mass by multiplying the initial value by 2(16). Levels of dd-cfDNA are high in the immediate post-transplant period because of the transplant surgery and then logarithmically decay in a two-phase kinetic fashion(10). Therefore, we excluded dd-cfDNA values prior to day 30 to avoid confounding from elevated levels in the early post-transplant period.

Statistical Analysis

Continuous variables were described using mean (standard deviation) or median (interquartile range), and categorical variables were summarized using counts (percentage). Nonparametric tests were used when indicated. Univariate comparisons were conducted using χ2 or Fisher exact tests for categoric variables and t tests for continuous variables. Multivariable analysis based on Cox regression models were used to assess the relationship of the time-dependent covariates of Molecular AMR and ISHLT Clinical AMR with the outcomes of CLAD and/or death. Since most subjects with Molecular or ISHLT Clinical AMR had only one episode during the study follow up, we treated Molecular and ISHLT Clinical AMR as a binary time-dependent covariate with an initial value of 0 from the time of transplant for all patients which was then recoded to 1 when AMR first occurred and remained as 1 thereafter(17, 18). Other time-invariant covariates included in the multivariable models were prespecified based on the potential for influencing the outcome of CLAD or death, and included recipient age, sex, single vs double lung transplant, PGD grade 3, LAS score, and transplant center. The proportional hazards assumptions (no time varying coefficients) for all Cox models were evaluated based on scaled Schoenfeld residuals and score tests and did not violate the proportional hazards assumption(17). The Little’s MCAR test was used to assess if data was missing completely at random (MCAR). To ameliorate the missing data the R package “mice” (Multivariate Imputation by Chained Equations) was used to impute the missing values of the covariates. Mice manages the uncertainty related to the missing values for data that are at least missing at random (MAR). The mice package uses other covariates in the data set to predict values of the missing data. Due to the pattern and randomness of the missing data, Markov Chain Monte Carlo (MCMC) was used to impute the data (Jakobsen et al. (2017). Analyses were performed using STATA 17, R version 4.4.2, GraphPad Prism 10.1.2 and SPSS 28. P values were 2-sided with significance indicated by a value < 0.05.

Results

Of the 305 subjects enrolled and transplanted at the time of our analysis, 273 subjects had undergone adjudication for AMR. We excluded 20 subjects that died/censored <90 days post-transplant, 4 with insufficient PFT data to adjudicate for CLAD due to death during the index hospitalization, 23 subjects without dd-cfDNA values > 30 days post-transplant, 6 subjects with missing DSA data, and 11 subjects with missing PFT data, leaving 209 subjects in the final analysis (Figure 1). The median (IQR) follow-up until death or censor for the cohort was 4.47 (2.91, 7.34) years. The mean (SD) age at transplantation was 55 (14) years, 47% were female, and 76% had a bilateral lung transplantation (Table 1). There was no significant difference in baseline demographic characteristics between the subjects excluded (and not adjudicated for AMR) vs. the subjects included in the final analysis (Supplementary Table 1). A Little’s MCAR test was consistent with missing data as missing completely at random (χ² = 36.1, p = 0.205). Sixty-five patients developed CLAD at a median of 580 (353, 771) days post-transplant.

Figure 1:

Flow diagram of the study including number of patients with and without Molecular AMR.

Table 1:

Demographic Characteristics

Characteristic	Overall N = 209	Patients with Molecular AMR N = 61	Patients without Molecular AMR N = 148	p
Age (years)	55 (14)	53 (15)	55 (14)	0.42
BMI (kg/m2)	25.31 (4.73)	25.26 (4.37)	25.32 (4.89)	0.93
Lung Allocation Score at Transplant	48.82 (17.89)	46.72 (16.10)	49.74 (18.60)	0.28
Female Sex	99 (47%)	34 (56%)	65 (44%)	0.12
Race
White	173 (83%)	52 (85%)	121 (82%)	0.81
African American	27 (13%)	7 (12%)	20 (14%)
Other	9 (4%)	2 (3%)	7 (5%)
Double Lung Transplant	158 (76%)	47 (77%)	111 (75%)	0.75
Native Lung Disease
Interstitial Lung Disease	98 (47%)	27 (44%)	71 (48%)	0.83
Cystic Fibrosis	34 (16%)	10 (16%)	24 (16%)
Chronic Obstructive Pulmonary	42 (20%)	13 (21%)	29 (20%)
Disease	6 (3%)	3 (5%)	3 (2%)
Idiopathic Pulmonary Hypertension	29 (14%)	8(13%)	21 (14%)
Other

Incidence of Clinical and Molecular AMR after transplantation

Forty-six (22%) subjects experienced an episode of ISHLT Clinical AMR at a median of 282 (95, 543) days post-transplant. In all, 13/46 (28%) of Clinical AMR subjects met criteria for probable Clinical AMR and 33/46 (72%) met criteria for possible AMR. Twelve Clinical AMR subjects had histopathological evidence of ACR at the time of diagnosis (5 with A1 grade, 5 with A2 grade, 1 with A3 grade and 1 with A4 grade). Four Clinical AMR subjects had histopathological evidence of Organizing Pneumonia at diagnosis, 3 subjects had histopathologic evidence of capillaritis and 4 subjects had positive C4d staining. Ten Clinical AMR patients had positive microbiology at the time of diagnosis (8 bacteria, 1 viral and 1 fungal).

Sixty-one (29%) subjects experienced an episode of molecular AMR at a median 193 (56, 400) days post-transplant. Out of these 61 Molecular AMR episodes, 23 (38%) met criteria for Clinical AMR at the time of Molecular AMR diagnosis, an additional 14 (23%) Molecular AMR patients went on to experience Clinical AMR at a mean of 173 days following the diagnosis of Molecular AMR, and 5 (8%) of Molecular AMR patients had prior episodes of Clinical AMR. In total, Molecular AMR captured 42/46 patients that experienced an episode of Clinical AMR. The remaining 19/61 (31%) Molecular AMR patients did not experience an episode of Clinical AMR (Figure 2) at any time. There was no difference in baseline demographic characteristics between subjects with Molecular AMR vs subjects without Molecular AMR (Table 1). Fifty-eight (95%) of the Molecular AMR events were associated with Class II antibodies, with only 3 cases (4%) characterized by solely Class I antibodies. Fifty-six (92%) of the Molecular AMR events were associated with de novo DSA (Table 2).

Figure 2:

Breakdown of the number Molecular AMR patients that also experienced ISHLT Clinical AMR in relation to their time of diagnosis of Molecular AMR.

Table 2:

DSA Characteristics of Molecular AMR (n=61)

Median (IQR) days to DSA Positivity	30 (7, 204)
HLA Class
Class I	3 (5%)
Class II	39 (64%)
Class I and II	19 (31%)
DQ DSA	48 (79%)
Multiple DSA	35 (57%)
De Novo DSA	56 (92%)
Strength
Low	38 (62%)
Flow	11 (18%)
Cytotoxic	9 (15%)
Missing	3 (5%)

Association of Clinical and Molecular AMR with CLAD and Death

Twenty-six (43%) of the Molecular AMR patients developed CLAD at a median 589 (354, 732) days after diagnosis. Twenty-four (39%) of the Molecular AMR patients died at a median 839 (491, 1273) days after diagnosis (the causes of death are presented in Supplementary Table 2). Multivariable, time-dependent Cox regression models demonstrate that the development of Molecular AMR is significantly associated with the composite outcome of CLAD/Death (HR 2.00, 95% CI: 1.18 – 3.38, p = 0.010) (Table 3). The results remain consistent when separately analyzing the cause-specific association of Molecular AMR with Death (HR 2.49, 95% CI: 1.43 – 4.35, p = 0.005) and CLAD (HR 2.46, 95% CI: 1.34 – 4.52, p = 0.004). The development of Clinical AMR is also associated with an increased risk of CLAD/Death (HR 2.36, 95% CI: 1.42 – 3.87, p = 0.001).

Table 3:

Association of Molecular AMR with composite outcome of CLAD/Death

Model*	Composite: CLAD/Death
	HR (95% CI)	p
Molecular AMR	HR 2.00, 95% CI: 1.18 – 3.38	0.010
Clinical AMR	HR 2.36, 95% CI: 1.42 – 3.87	0.001
Molecular AMR w/o Clinical AMR	HR 2.14, 95% CI: 1.10 – 4.19	0.026
Molecular AMR w/o Clinical AMR, ACR or Pathogens	HR 2.45, 95% CI: 1.01 – 5.94	0.047

^*Multivariable analysis was adjusted for patients’ age, sex, single vs double lung transplant, PGD grade 3, LAS score, and transplant center.

To clarify if the relationship between Molecular AMR and CLAD/Death was driven predominantly by the presence of molecular allograft injury alone (instead of simply the combination of DSA + molecular allograft injury), we performed a sensitivity analysis including all subjects in the cohort that experienced a dd-cfDNA >1% after 30 days post-transplant (n = 158). The results remained consistent in that Molecular AMR was still associated with an increased risk of CLAD/Death (HR: 1.95, 95% CI 1.12 – 3.29, p = 0.013), indicating that the association of Molecular AMR with CLAD/Death is not due to the presence of allograft injury on the molecular level alone (dd-cfDNA > 1%), but the combination of DSA + dd-cfDNA.

Given that the presence of Clinical AMR at the time of, or prior to, diagnosis of Molecular AMR may influence the association of Molecular AMR with CLAD/Death, we performed a sensitivity analysis excluding the 23 Molecular AMR events that had concomitant Clinical AMR at diagnosis as well as the 5 Molecular AMR events that had prior episodes of Clinical AMR. The results remain consistent in that Molecular AMR is significantly associated with CLAD/Death (HR 2.14, 95% CI: 1.10 – 4.19, p = 0.026).

Further, because episodes of ACR or positive microbiology at Molecular AMR diagnosis may influence levels of dd-cfDNA and downstream clinical outcomes, we performed a sensitivity analysis excluding Molecular AMR without Clinical AMR (n=23) or concomitant ACR/pathogen (n=10). In this most stringent model, the results remain consistent in that Molecular AMR was significantly associated with CLAD/Death (HR 2.45, 95% CI: 1.01 – 5.94, p = 0.047). The results also stand when imputing missing values using multivariate imputation with chained equations (Supplementary Table 3).

Finally, because the association of Molecular AMR with CLAD/Death may be driven solely by the presence of DSA alone (regardless of the level of dd-cfDNA), we performed a sensitivity analysis adjusting for the presence of DSA. The results remained consistent in that Molecular AMR was significantly associated with CLAD/Death (HR 2.05, 95% CI: 1.05 – 4.03, p = 0.037). Although a subject may have DSA without the development of Molecular AMR, we performed a correlation matrix analysis of regression coefficients between Molecular AMR and DSA positivity to assess for collinearity between the two variables which indicated a low correlation (0.19) with no substantial collinearity between the variables.

Discussion

In this analysis of two multicenter prospective cohort studies, we propose a definition of Molecular AMR by combining the presence of DSA with elevated molecular markers of allograft injury -assessed by plasma dd-cfDNA - and demonstrate that the development of Molecular AMR is significantly associated with CLAD and Death. In this manner, we identify a novel at-risk population of patients potentially experiencing significant allograft injury on the molecular level that is not captured by the current ISHLT definition of AMR. These findings have unique implications for the way allograft injury is assessed in a patient at risk for AMR through criteria predicated on molecular evidence of allograft injury and are in accordance with the 2016 ISHLT consensus statement on AMR calling for further investigation into molecular diagnostic criteria for AMR as a research priority.

Our lab has previously reported that dd-cfDNA may increase several weeks-to-months prior to the clinical manifestations of acute rejection(3). Consistent with this phenomenon, our data demonstrate that over a quarter of ISHLT Clinical AMR subjects experienced elevations of dd-cfDNA in the setting of Molecular AMR on an average of 6 months prior to diagnosis of Clinical AMR. Further, an additional 42% of Molecular AMR patients did not develop downstream clinical manifestations of Clinical AMR over study follow-up. These results suggest a sub-clinical impact of DSA on allograft health defined by the degree of allograft injury on the molecular level and expand on prior studies associating molecular allograft injury with the development of CLAD and Death(9, 13). Our findings also expand on the evolving role of identifying high-risk DSA characteristics that may benefit from pre-emptive treatment to prevent downstream allograft dysfunction. This is consistent with prior data identifying specific DSA characteristics, such as complement-binding capacity and DSA persistence, with an increased risk of CLAD(19). We extend these principles and provide rationale for the continued investigation into the role of serial monitoring of molecular evidence of allograft injury for routine surveillance in the post-transplant period.

Despite the recognized shortcomings of the current ISHLT definitions of AMR, the 2016 working group indeed provided clear diagnostic criteria based on the available evidence, which importantly, carry prognostic significance(1, 2). Therefore, the current ISHLT definition does provide needed value in the identification and risk-stratification of this vulnerable population of lung transplant recipients. As such, the conceptualization of Molecular AMR that we propose may serve as an adjunct to existing definitions by identifying an additional group of at-risk lung transplant recipients vulnerable to AMR. The current ISHLT consensus report acknowledges that a key challenge in the diagnosis and grading of AMR in lung transplantation is the “variable relationship between DSA and the presence of graft damage and dysfunction”(2). Accordingly, a definition of Sub-clinical AMR was put forth and composed of the presence of DSA/Lung Histology and C4d staining. Our findings directly address ISHLT research priorities emphasizing the need for molecular diagnostic criteria for pulmonary AMR, and suggest that molecular evidence of allograft dysfunction, as assessed by dd-cfDNA, may serve as a valuable tool in the diagnosis of Sub-clinical AMR.

This study has several limitations. As an observational study, residual confounding may exist despite adjusting for several confounding variables and performing numerous sensitivity analyses in our multivariable models. Although adjusted for in our models, center-level differences regarding treatment practices for DSA and post-transplant monitoring strategies may further contribute to residual confounding, even despite no statistically significant difference in transplant center between the two groups. Adequate DSA, CLAD and dd-cfDNA data were missing in about 15% of our cohort and excluded according to our exclusion criteria, and although we considered this data missing at random (as supported by the Little’s MCAR test), this may contribute to selection bias. However, our results stand when imputing the missing data using multiple imputations with chained equations. Although the sample size of our study was robust, these findings require validation in separate, larger multicenter trials. Our definition does account for cases of potential DSA negative rejection or rejection secondary to non-HLA antibodies. Further, despite the nomenclature of Molecular AMR, we cannot be certain that this process is mediated by antibody-dependent cell-mediated cytotoxicity or complement activation – which is the current conceptualization of the pathophysiologic mechanisms of AMR. However, the current ISHLT AMR definition also does not capture the presence of these pathophysiologic processes with certainty. Given these limitations, our results should be viewed as hypothesis generating and support further research into the role of dd-cfDNA for the diagnosis of AMR and risk stratification of DSA positive lung transplant recipients. Future studies should also evaluate the significance of alternative threshold values of dd-cfDNA for Molecular AMR, and ultimately, the impact of treatment of Molecular AMR on clinically relevant outcomes in randomized controlled trials. Finally, we can’t ignore the complexities of navigating a diagnosis of AMR that incorporates elevated dd-cfDNA and positive DSA in the context of positive infection or microbiology, which may also increase dd-cfDNA levels. While our sensitivity analysis excluding Molecular AMR patients with concomitant infection yielded similar findings as our primary analysis, future studies should aim to further explore this interaction and validate our results.

In conclusion, the development of Molecular AMR – defined as the presence of DSA with elevated dd-cfDNA, is associated with the development of CLAD and death. Molecular AMR may identify a population of lung transplant recipients at risk of CLAD/Death that are not captured by current ISHLT AMR definitions. This study provides rationale for further investigation into the role of molecular markers of allograft injury in the diagnosis of AMR.