Follistatin and Soluble Endoglin Predict 1-year Non-Relapse Mortality after Allogeneic Hematopoietic Cell Transplant

Abstract

Damage-associated angiogenic factors (AF), including follistatin (FS) and soluble endoglin (sEng), are elevated in circulation at the onset of acute graft-versus-host disease (GVHD). We hypothesized that regimen-related tissue injury might also be associated with aberrant AF levels, and sought to determine the relevance of these AF on non-relapse mortality (NRM) in patients with and without acute GVHD. To test our hypothesis, we analyzed circulating levels of FS, sEng, angiopoietin-2 (Ang2), epidermal growth factor (EGF), vascular endothelial growth factor (VEGF) -A and -B, placental growth factor (PlGF), and soluble VEGF receptor [sVEGFR]-1 and -2, in plasma samples from patients enrolled on Blood and Marrow Transplant Clinical Trials Network (BMT CTN) 0402 (n=221), which tested GVHD prophylaxis after myeloablative hematopoietic stem cell transplantation (HCT). We found that the interaction between FS and sEng had an additive effect in their association with 1-year NRM. In multivariate analysis, patients with the highest levels of day +28 FS and sEng had a 14.9-fold hazard ratio (HR) of NRM (95% confidence interval 3.2–69.4, p<0.01) when compared to low levels of FS and sEng. We validated these findings using an external cohort of patients (n=106). Pre-HCT measurements of FS and sEng were not associated with NRM, suggesting that elevations in these factors early post-HCT may be consequences of early regimen-related toxicity. Determining the mechanisms responsible for patient-specific vulnerability to treatment toxicities and endothelial damage associated with specific AF elevation may guide interventions to reduce NRM post-HCT.

Introduction

Allogeneic hematopoietic stem cell transplantation (HCT) can be curative for patients with high-risk or relapsed hematologic malignancies. While advances in supportive care have improved HCT outcomes^{1, 2}, acute graft-versus-host disease (GVHD) and non-relapse mortality (NRM) due to organ toxicity remain common life-threatening post-transplant complications^3–6. Strategies to improve HCT and GVHD-specific outcomes include recent clinical trials aimed at determining optimal prophylaxis^7–10, as well as treatment^{11, 12}, of acute GVHD. While traditionally thought to be solely related to organ-specific tissue damage mediated by activated allo-reactive T-cells¹³, increasing evidence has demonstrated an important role for endothelial and microvasculature damage in the development of GVHD^14–22.

The process of angiogenesis, as a hallmark of wound healing, has been well-described, with a tightly regulated diverse repertoire of pro- and anti-angiogenic factors working in concert to repair damaged tissues^23–25. Angiogenic stimuli are also linked with inflammatory responses through the induction of growth factors, chemokines, and cytokines^26–28. Inflammation and angiogenesis may thus be linked biologic processes influencing the onset of and recovery from organ damage and GVHD after allogeneic HCT. Indeed, accumulating evidence has suggested a dysregulation in the balance of angiogenic factors (AFs) which promote tissue healing and wound repair, versus those factors yielding tissue damage and inflammation, in the setting of acute GVHD^29–33.

We recently identified a panel of 5 circulating AFs which are altered in patients developing acute GVHD³⁴, using serum and plasma samples from patients enrolled in Blood and Marrow Transplant Clinical Trials Network (BMT CTN) trials 0302¹² and 0802¹¹. At the onset of acute GVHD, levels of AFs associated with tissue healing/repair (i.e. epidermal growth factor [EGF] and vascular endothelial growth factor A [VEGF-A]) were low, while levels of AFs indicating tissue damage/inflammation (i.e. follistatin [FS], soluble endoglin [sEng], and placental growth factor [PlGF]) were elevated, compared to HCT recipients without GVHD. Additionally, persistently elevated FS was an independent predictor of mortality, and persistently elevated sEng was associated with steroid resistant acute GVHD³⁴. However, the impact of these AFs on other (non-GVHD) outcomes, and the association between clinical endpoints and longitudinal trajectories of AF levels remain unknown.

We hypothesized that a pattern of tissue damage, manifested by elevated levels of inflammation- and damage-associated AFs early after HCT, would be associated with increased NRM with or without subsequent acute GVHD. We analyzed plasma samples collected at day +28 after HCT, an important time point after myeloablative conditioning³⁵ and an array of regimen-related toxicities³⁶. We investigated any interaction between AFs in relation to NRM, hypothesizing that the dysregulation of multiple AFs likely contributes to these adverse post-HCT outcomes.

Materials and Methods

Angiogenic Factor Levels

We studied plasma samples from patients enrolled in BMT CTN acute GVHD prophylaxis study 0402, a randomized trial of tacrolimus/methotrexate versus tacrolimus/sirolimus, after myeloablative matched sibling peripheral blood stem cell HCT. Plasma concentrations of AF (angiopoietin-2 [Ang2], EGF, FS, VEGF-A and -B, sEng, PlGF, and soluble VEGF receptor [sVEGFR]-1 and -2) collected at day 0 and day +28 post-HCT were quantified by MILLIPLEX magnetic bead panels (Millipore, Billerica, MA). Plasma AF levels were subsequently quantified in an independent external validation cohort (n=106) using samples collected at day +28 post-HCT from patients undergoing HCT at the University of Minnesota.

Statistical Methods

We analyzed AFs to determine their association with the cumulative incidence of NRM, treating relapse as a competing risk. The log transformation was always used on AFs in the analyses. Patients with day +28 samples who survived to day +28 without relapse or disease progression were included (221 of 304 enrolled in the trial). Given the number of AF evaluated and thus multiple comparisons, the false-discovery rate was used to adjust for the type-I error in univariate assessments of each of the markers on one-year NRM.

Fine and Gray regression analysis was used to examine the independent effect of AFs on NRM treating AFs as continuous factors. Interactions were also evaluated. For analysis on the endpoint of NRM, acute GVHD was defined as time-dependent grade II–IV acute GVHD occurring between day 0–100 post-HCT, with grade defined as maximum grade occurring during that time period. Other potential confounding factors tested included gender (male vs. female), age, disease (acute myelogenous leukemia [AML], acute lymphoblastic leukemia [ALL], chronic myelogenous leukemia [CML], myelodysplastic syndrome [MDS]), disease status (1^st complete remission [CR1], 2^nd complete remission [CR2], vs. other), conditioning (cyclophosphamide/total body irradiation [Cy/TBI] versus etoposide/TBI), the randomized GVHD prophylaxis (tacrolimus/methotrexate [Tac/MTX] vs. tacrolimus/sirolimus [Tac/Siro]), recipient CMV serostatus (negative vs. positive), and year of transplant (2006–2008 vs. 2009–2011).

We also examined optimal cut-points for two of the AFs that showed an association. The optimal cut-points were selected based on the maximally selected log-rank statistics for cut-points between the 10% and 90% quantile using the upper bound of the P-value as described by Hothorn and Lausen in the maxstat package from R³⁷. This adjusts for the problem of multiple testing. The bootstrap resampling technique was applied to estimate the 95% confidence intervals (CI) for the optimal cut-points. The 2.5 and 97.5 percentiles of 1000 bootstrap samples were cited around the estimation method.

We validated these findings using an independent external cohort of patients (n=106) from the University of Minnesota. Visual comparison of cumulative incidence curves are presented for suggestive interpretation of discrimination and calibration. Somer’s D_XY was also calculated using the val.surv function from the RMS package in R to more formally cite discrimination. Incidence and grade of acute GVHD by day 100, chronic GVHD, and cause of death, were adjudicated by the BMT CTN 0402 Protocol Committee⁷. P-values were two-sided. SAS 9.4 (SAS Institute, Cary, NC) and R3.6.0 were used for all statistical analyses.

Results

Patient and Transplant Characteristics

Two hundred and twenty-one patients from BMT CTN 0402⁷ had day +28 samples available and were included on study. Patient and donor characteristics are summarized in Table 1. The median patient age was 43 years (range 12–59 years), and median donor age was 43 years (range 13–66 years). One hundred and sixteen (53%) patients and 132 (60%) donors were males. All patients were transplanted for a hematologic malignancy, the majority acute leukemia: AML in 99 patients (45%), and ALL in 88 patients (40%). Per the BMT CTN 0402 protocol, all patients received myeloablative conditioning, with either Cy/TBI (n=174, 79%), or etoposide/TBI (n=47, 21%).

Table 1:

Patient and Transplant Characteristics

Characteristic	Patients
No. of Patients	221
Patient age in years
<18 years old	1
Median (range), (IQR)	43 (12–59), (35–49)
Patient gender
Male, No. (%)	116 (52%)
Disease, No. (%)
Acute leukemia	188 (85%)
Chronic myeloid leukemia (CML)	17 (8%)
Myelodysplastic syndrome (MDS) / Myeloproliferative Disease (MPD)	16 (7%)
Disease stage, No. (%)
Acute Leukemia: CR1	158 (71%)
Acute Leukemia: CR2	30 (14%)
CML: AP	2 (1%)
CML: CP	15 (7%)
Chronic myelomonocytic leukemia (CMML)	3 (1%)
MDS*	13 (6%)
Graft source
HLA-matched sibling donor PBSC	221 (100%)
Conditioning regimen, No. (%)
Cy/TBI	174 (79%)
Etoposide/TBI	47 (21%)
GVHD prophylaxis, No. (%)
Tacrolimus/Methotrexate	113 (51%)
Tacrolimus/Sirolimus	108 (49%)
CMV serostatus
Recipient seropositive	140 (63%)
Year of transplant
2006 – 2008	65 (29%)
2009 – 2011	156 (71%)
Cumulative incidence of acute GVHD grade II–IV	31% (95% CI, 25–37%)
Cumulative incidence of acute GVHD grade III–IV	11% (95% CI, 7–15%)
Cumulative incidence of chronic GVHD at 1-year	48% (95% CI, 41–55%)

Abbreviations: IQR, interquartile range; CR1, first complete remission; CR2, second complete remission; AP, accelerated phase; CP, chronic phase.

^*Includes MDS with isolated del(5q) (n=1), MDS unclassified (n=1), refractory anemia (n=1), refractory anemia with ringed sideroblasts (n=1), cytopenia with multilineage dysplasia (n=1), cytopenia with multilineage dysplasia and ringed sideroblasts (n=1), refractory anemia with 5–10% blasts (n=3), and refractory anemia with 10–20% blasts (n=4).

Transplant Outcomes

The 1-year overall survival (OS) was 78% (95% CI, 73–83%), and the cumulative incidence of NRM at 1-year was 11% (95% CI, 7–15%). Grade II–IV acute GVHD developed in 31% (95% CI, 25–37%) of the study patients, and grade III–IV acute GVHD in 11% (95% CI, 7–15%). The cumulative incidence of chronic GVHD at 1-year was 48% (95% CI, 41–55%).

Associations between Day +28 Angiogenic Factors and 1-year NRM

Twenty-five patients experienced 1-year NRM, our primary endpoint. In univariate analysis, increased levels of day +28 sEng and FS were each significantly associated with 1-year NRM, Table 2. There was no association with NRM for either EGF or VEGF-A (AFs associated with tissue repair), nor with other AFs reflective of tissue damage/inflammation (Ang-2, PlGF, sVEGFR-1 or -2, or VEGF-B), Table 2. We also found no impact of pre-HCT levels of FS and sEng on 1-year NRM (Supplementary Table S1).

Table 2:

1-year NRM by Plasma AF Levels, Univariate Analysis

All Patients (n=221)	HR of NRM* (per doubling)	95% CI	Adjusted P-value
sEng	2.3	1.2–4.3	0.05
FS	1.9	1.2–2.9	0.01
Ang-2	1.5	1.0–2.1	0.09
PIGF	1.6	0.8–2.9	0.29
sVEGFR-1	1.1	0.9–1.3	0.14
EGF	1.0	0.8–1.1	0.6
VEGF-A	1.1	0.8–1.5	0.53
sVEGFR-2	1.2	0.7–2.0	0.53
VEGF-B	0.8	0.4–1.9	0.66

^*Hazard ratio of NRM per doubling of AFs.

Treating the logarithm of FS and the logarithm of sEng as both continuous factors and binary (based on optimal cut-points), we came to the same conclusion that the combination of FS and sEng showed an independent association with NRM. The first model measured FS, sEng, and an interaction term as continuous factors. Higher concentrations of each factor were associated with the highest risk of NRM with a statistically significant positive interaction between the two biomarkers (Table 3a). The effect of FS also showed an increasing impact on NRM within each increase in quartile of sEng (Supplementary Table S2). After adjustment for age and acute GVHD, the optimal cut-points were estimated to be 6.98 (95% CI, 6.77–8.08), and 6.65 (95% CI, 5.92–6.67) for FS and sEng, respectively. The interaction of these factors based on these cut-points was also associated with higher risk of NRM: FS<6.98 and sEng<6.65, 3% (0–7%) CI of NRM; FS<6.98 and sEng≥6.65, 11% (5–16%) CI of NRM; FS≥6.98 and sEng<6.65, 0% CI of NRM; FS≥6.98 and sEng≥6.65, 40% (20–60%) CI of NRM; (adjusted p<0.01, Figure 1). In regression analysis using the cut-points, patients with the highest levels of day +28 FS and sEng (n=32) had a 14.9-fold higher hazard (HR) of NRM (95% CI, 3.2–69.4, p<0.01, Table 3b). Older patient age was also independently associated with 1-year NRM (HR 1.8 per decade, 95% CI 1.2–2.6, p<0.01). Neither patient gender (p=0.74), disease status (p=0.79), CMV serostatus (p=0.10), GVHD prophylaxis (Tac/MTX versus Tac/Siro, p=0.74), or the development of grade II–IV acute GVHD (p=0.11) independently influenced risks of NRM or showed a confounding effect on the biomarkers and NRM.

Table 3a:

Multiple Regression Analysis for 1-year Non-Relapse Mortality

All Patients (n=221)	HR of NRM* (per doubling)	95% CI	P-value
sEng	0.1*	0.04–0.5	<0.01
FS	0.1*	0.03–0.4	<0.01
Interaction of sEng and FS	1.3	1.1–1.5	<0.01
Age (/decade)	1.9	1.2–2.9	0.01

Model controlled for acute GVHD (p=0.06).

^*Hazard ratio of NRM per doubling of sEng and FS, including interaction between sEng and FS. Values <1 given the interaction term between sEng and FS: as sEng increases, there is a stronger impact of FS on NRM and vice versa (stronger impact of sEng on NRM with increasing FS).

Figure 1: Non-Relapse Mortality by FS-sEng.

After adjustment for age and acute GVHD, higher FS and sEng cut-points were associated with a greater risk of NRM.

Table 3b:

Multiple Regression Analysis for 1-year Non-Relapse Mortality

Factor	N	RR (95% CI)	P-value
Optimal Cut-Points			<0.01
FS<6.98, sEng<6.65*	67	1.0
FS<6.98, sEng≥6.65	122	3.7 (0.8–16.4)	0.09
FS≥6.98, sEng<6.65	7	0	<0.01
FS≥6.98, sEng≥6.65	25	14.9 (3.2–69.4)	<0.01
Age (/decade)	221	1.8 (1.2–2.6)	<0.01

^*Reference group.

Associations between FS-sEng Composite Score and GVHD

There was no association of grade II–IV acute GVHD or chronic GHVD with the FS-sEng cut-points (Tables 4 and 5, Supplementary Figure S1).

Table 4:

Multiple Regression Analysis of Grade II–IV Acute GVHD

Factor		N	HR (95% CI)	P-value
Optimal Cut-Points (FS/sEng)				0.83
FS<6.98, sEng<6.65*		67	1.0
FS<6.98, sEng≥6.65		122	1.1 (0.7–1.9)	0.65
FS≥6.98, sEng<6.65		7	0.5 (0.1–3.8)	0.47
FS≥6.98, sEng≥6.65		25	1.0 (0.4–2.4)	0.99
GVHD Prophylaxis
Tac/MTX*		113	1.0
Tac/Siro		108	0.7 (0.5–1.2)	0.22

^*Reference group.

Table 5:

Multiple Regression Analysis for 1-year Chronic GVHD

Factor		N	HR (95% CI)	P-value
Optimal Cut-Points (FS/sEng)				0.08
FS<6.98, sEng<6.65*		67	1.0
FS<6.98, sEng≥6.65		122	0.8 (0.5–1.3)	0.36
FS≥6.98, sEng<6.65		7	1.5 (0.7–3.4)	0.32
FS≥6.98, sEng≥6.65		25	0.4 (0.2–0.9)	0.03
GVHD Prophylaxis
Tac/MTX*		113	1.0
Tac/Siro		108	1.5 (1.0–2.1)	0.053

^*Reference group.

Associations between FS-sEng and 1-year overall survival

In cox regression analysis, a higher FS and higher sEng was associated with a 4.4-fold increased risk of mortality (Table 6, Supplementary Figure S2). The competing risk of relapse was similar across all 4 groups (data not shown).

Table 6:

Cox Regression Analysis for 1-year Overall Survival

Factor		N	HR (95% CI)	P-value
Optimal Cut-Points (FS/sEng)				<0.01
FS<6.98, sEng<6.65*		67	1.0
FS<6.98, sEng≥6.65		122	1.4 (0.7–2.9)	0.36
FS≥6.98, sEng<6.65		7	0.9 (0.1–6.8)	0.87
FS≥6.98, sEng≥6.65		25	4.4 (1.9–10.1)	<0.01
Age (/decade)		221	1.4 (1.0–1.9)	0.04

^*Reference group.

Cause of NRM

Deaths due to organ toxicity (e.g. liver failure/veno-occlusive disease [VOD], respiratory failure, infection, or other organ toxicity) were more frequent with a higher FS and higher sEng (Supplementary Figure S3).

Validation Cohort

Patient and transplant characteristics of the independent external validation cohort, which included recipients of both myeloablative and non-myeloablative conditioning, are shown (Supplementary Table S3). In the validation cohort, the overall day +28 FS-sEng combination showed a significant association with NRM (p=0.02, Table 7).

Table 7:

External Model Validation Cohort: Univariate Analysis for 1-year Non-Relapse Mortality

Factor	N	HR (95% CI)	P-value
Optimal Cut-Points (FS/sEng)			0.02
FS<6.98, sEng<6.65*	12	0%
FS<6.98, sEng≥6.65	45	16% (5–26%)
FS≥6.98, sEng<6.65	8	0%
FS≥6.98, sEng≥6.65	41	27% (13–41%)

^*Reference group.

Supplementary Figure S4 shows that the curves are separated showing discrimination. Testing FS-sEng in this cohort showed a satisfactory performance (Somer’s D_XY = 0.353). All curves in the validation dataset do not agree perfectly with those in the 0402 dataset which suggests there appears to be some miscalibration – but they still trend in a similar direction. We believe these results were consistent with our results from the primary BMT CTN 0402 analysis although showing potentially less interaction between FS and sEng.

Discussion

The interplay between AFs associated with tissue healing and repair versus those associated with tissue damage and inflammation is a complex and dynamic process. In this report, we show that levels of two specific AFs, sEng and FS, are associated with NRM after HCT, independent of acute GVHD. In addition, we showed that there was positive additive interaction between day +28 FS and sEng on 1-year NRM. We have confirmed this association of the day +28 FS-sEng interaction in an independent external validation cohort. Pre-HCT FS and sEng levels had no significant association with 1-year NRM, suggesting that elevations in these factors at day +28 reflect regimen-related toxicity and unresolved tissue damage, which may then predispose patients to NRM.

It remains unclear whether the increased plasma levels of FS and sEng contribute to NRM directly, or whether they are markers of the underlying tissue injury. Increased circulating levels of FS are also seen in sepsis^{38, 39} and in response to post-surgical stress⁴⁰. Importantly, FS can be induced by both infectious and inflammatory stimuli^41–43. FS is secreted by hepatocytes and by circulating endothelial cells, but only in endothelial cells is FS expression induced by inflammatory cytokines^{44, 45}, supporting our hypothesis that elevated levels of these AFs may reflect endothelial damage. FS is best known as a potent antagonist of activin-A, a member of the transforming growth factor-beta (TGF-β) superfamily⁴⁶, and can bind and neutralize the activity of other members of the TGF-β superfamily (e.g. bone morphogenetic proteins [BMPs], myostatin)⁴⁷. Persistently elevated FS may reflect increased tissue damage and ongoing FS-induced angiogenic stimuli.

Endoglin, also known as CD105, is expressed by endothelial cells during neoangiogenesis and is a co-receptor for TGF-β superfamily members⁴⁸. Expression of endoglin is upregulated and persistent in response to inflammatory conditions⁴⁹. Membrane shedding of endoglin yields sEng, which can be further induced by tumor necrosis factor-alpha (TNF) or other cytokines⁵⁰. Circulating sEng sequesters TGF-β family ligands and inhibits ligand-receptor binding at the endothelial cell surface⁵¹, thus indirectly serving as a TGF-β antagonist⁵². Additional mechanistic studies are necessary to determine how FS and sEng modulate post-HCT outcomes.

The identification of circulating biomarkers predictive of post-HCT outcomes may lead to novel therapeutic strategies designed to modulate the risk of developing, and outcome after treatment of, complications such as GVHD and NRM. Several studies have provided evidence correlating select biomarkers with specific HCT complications, including suppression of tumorigenicity-2 (ST2) with acute GVHD and particularly treatment-resistant acute GVHD⁵³ and NRM^{21, 53, 54}; T-cell immunoglobulin mucin-3 (TIM-3) with severe acute GVHD^{21, 55}; regenerating islet-derived protein 3 alpha (Reg3α) with gastrointestinal GVHD^{54, 56, 57}; and chemokine (C-X-C motif) ligand 9 with chronic GVHD^58–60. Recently, Abu Zaid et al. reported a correlation between elevated day +28 plasma ST2 and TIM3 levels with 2-year NRM and OS, in samples also from BMT CTN 0402⁶¹. Significantly elevated ST2 has also been reported in pre-eclampsia, thought to be secondary to systemic vascular inflammation⁶², further strengthening the potential parallel between endothelial dysfunction and inflammation.

The number of patients experiencing NRM in the BMT CTN 0402 cohort was relatively low (<12%), which may have limited the strength of our analysis. Additionally, the homogenous patient population in our study is both a strength and a weakness. Our cohort of samples from BMT CTN 0402 only included patients receiving myeloablative TBI-based conditioning and HLA-matched sibling donors, with either tacrolimus/methotrexate or tacrolimus/sirolimus for GVHD prophylaxis, allowing relatively controlled clinical HCT variables, as previously mentioned by Abu Zaid et al.⁶¹. However, the impact of FS and sEng with other conditioning intensity or graft sources was thus unexplored. We attempted to overcome this limitation using our validation cohort, which included a more heterogeneous HCT population; the day +28 FS-sEng score retained its impact on NRM.

Our analysis uniquely tested several AFs and identified a pattern of additive positive interaction between elevated day +28 FS and sEng and the development of 1-year NRM. Further clarification of the underlying mechanisms responsible for the impact of FS and sEng on NRM will be valuable for future efforts designed to translate these results into clinically meaningful assays and novel interventions to limit NRM after allogeneic HCT.