A panel of ST2, Reg3α, Elafin, and TNFR1: Predictive value for acute GvHD and outcomes in hematological malignancy patients post allo-HSCT

Abstract

Allogeneic hematopoietic cell transplantation (allo-HSCT) is an essential therapeutic modality hematological malignancies, but acute graft-versus-host disease (aGvHD) persists as a leading cause of non-relapse mortality (NRM) . Cytokine biomarkers have already been used to predict aGvHD and outcomes. However, the standard guidelines for aGvHD biomarker panels remain controversial.

We retrospectively analyzed the association of a biomarker panel (suppressor of tumorigenesis 2 [ST2], regenerating islet-derived 3 α [REG3α], Elafin, and tumor necrosis factor 1 [TNFR1]) in serum with the onset of 100-day aGvHD, 12-month NRM, and overall survival (OS) in 141 hematological malignancies patients at 19 ± 5 days after allo-HSCT from January 2022 to August 2023. Multivariable analysis showed that ST2 (P < .001) were strongly correlated with aGvHD, and TNFR1 was significantly associated with 12-month NRM and OS (P < .001). The panel of ST2, REG3α, Elafin, and TNFR1 demonstrated the best performance in diagnosis of 100-day aGvHD (area under the curve [AUC] = 0.79) and in the prediction of 12-month NRM (AUC = 0.74) and OS (AUC = 0.71). The 4-biomarker panel’s risk classification predicted the 12-month cumulative incidence of NRM (43% vs 11%, P < .001) and 12-month OS (51% vs 82%, P < .001) for the high-risk and low-risk groups, respectively. Our results suggest that a combination of ST2, REG3α, Elafin, and TNFR1 is an excellent biomarker predictive panel for aGvHD diagnosis and outcomes after allo-HSCT.

1. Introduction

Allogeneic hematopoietic stem cell transplantation (allo-HSCT) is a well-established curative therapy for various malignant and nonmalignant hematological diseases. However, the incidence of acute graft-versus-host disease (aGvHD) is around 30–50% and the standard first-line therapy remains steroids, with a response rate of about 50%.^[1–3] A classic aGvHD occurs within 100 days after allo-HSCT and the median onset of aGvHD is approximately 1 month after transplant. It major affects the skin, liver, and gastrointestinal (GI) tract.^[2] Currently, aGvHD diagnosis mainly depends on clinical manifestations and target organ pathological biopsy, lacking disease specificity and clinical feasibility.^[4] Moreover, clinical symptoms at onset is poor to predict long-term outcomes.^[5]

Numerous biomarkers have been discovered to aid in the early diagnosis and accurate risk stratification of aGvHD. Some of these biomarkers correlate with immunologic activation and systemic inflammation, such as interleukin(IL)-6, IL-8, soluble IL-2 receptor and tumor necrosis factor 1 (TNFR1).^[6–9] Other biomarkers reflect damage by aGvHD to target organs, such as Elafin, a biomarker for skin aGvHD, regenerating islet-derived 3 α (REG3α) for GI aGvHD, and suppressor of tumorigenesis 2 (ST2) for systemic aGvHD.^[10–15] Combinations of biomarkers can predict outcomes better than individual biomarkers. Several groups have validated aGvHD biomarker combination to predict outcomes at diagnosis. The Mount Sinai aGvHD International Consortium (MAGIC) algorithm probability uses the concentrations of 2 biomarkers, ST2 and REG3α, to predict the response to systemic therapy, risk of non-relapse mortality (NRM), and survival.^[16,17] Third Military Medical University reported a combination of toll-like receptor 4, TNFR1, transforming growth factor-beta, and Elafin could be a new 4-biomarker panel to assist aGVHD diagnosis, grading, and evaluation of steroid sensitivity for clinical aGVHD patients.^[18]

However, the clinical practice standards for the diagnosis and prognosis of aGvHD remain controversial to date. In this study, we evaluated ST2, REG3α, Elafin and TNFR1 at 19 ± 5 days in serum after allo-HSCT to predict the risk of 100-day aGvHD, 12-month NRM and overall survival (OS) after allo-HSCT.

2. Methods

2.1. Study design and patients

This was a single-center retrospective study, Figure 1 provides the flow chart of this study. Inclusion criteria: patients completed serum aGvHD biomarker panel detection at 19 ± 5 days after allo-HSCT; had complete clinical data and follow-up data; grade II to IV aGvHD diagnosis was consistent international consensus diagnostic criteria (MAGIC criteria^[4]); hematological malignancy patients who underwent allo-HSCT in the department of hematology, Aerospace Center Hospital. Exclusion criteria: late aGvHD; chronic graft-versus-host disease; corticosteroid contraindications; donor lymphocyte infusion related aGvHD. A total of 141 patients who received allo-HSCT between January 2022 and August 2023 were enrolled, with follow-up until August 2024. All of the 141 patients’ serum samples were collected at 19 ± 5 daysafter allo-HSCT. The study complied with the Declaration of Helsinki, was approved by the Aerospace Center Hospital Ethics Committee (no. 2024037-02), and all patients provided informed consent.

Figure 1.

The flow chart of this study. Complete data included serum levels of ST2, REG3α, Elafin, and TNFR1 at 19 ± 5 days after allo-HSCT, as well as 12-month clinical follow-up data after allo-HSCT. aGvHD = acute graft-versus-host disease, allo-HSCT = allogeneic hematopoietic cell transplantation, REG3α = regenerating islet-derived 3 α, ST2 = suppressor of tumorigenesis 2, TNFR1 = tumor necrosis factor 1.

2.2. aGvHD prophylaxis and therapy

All patients received myeloablative conditioning with modified busulfan, cyclophosphamide , and antithymocyte globulin. Graft-versus-host disease (GvHD) prophylaxis included cyclosporin A, mycophenolate mofetil, and short-term methotrexate. Treatment initiated with 1–2 mg/kg/day glucocorticoids, tapered after aGvHD symptom improvement. Basiliximab, ruxolitinib, antithymocyte globulin , and anti-CD25 monoclonal antibody served as second-line therapy for steroid-refractory cases.

2.3. Biomarker measurement and follow-up

Elafin/REG3α/ST2/TNFR1 were detected by multiparameter flow cytometry DxFLEX (Beckman Coulter Life Sciences Indianapolis, IN) in the Aerospace Center Hospital hematology laboratory. Human custom kits were purchased from Quantobio biotechnology company (Beijing, China), and performed according to manufacturer’s protocol. Serum samples were diluted 1:2 and measured at 19 ± 5 days post allo-HSCT via bead-based multiplex flow cytometry, with cytokines expressed in pg/mL. All biomarker values were log-10 transformed for use in algorithms. All patients completed a follow-up period of 12 months, and minimal residual disease was assessed monthly for 6 months post allo-HSCT, then quarterly.

2.4. Outcome ascertainment

aGvHD was diagnosed and graded by the “MAGIC criteria,”^[4] primarily assessing the extent of involvement of target organs such as skin, gastrointestinal tract (GI), and liver. NRM was defined as death from causes other than relapse. OS was defined as the time from enrollment to death from any cause.

2.5. Statistical methods

Patient characteristics between the aGvHD and non-aGvHDgroups were compared using the χ² or Wilcoxon 2-sample tests as appropriate. All biomarker values were log-transformed, and concentrations from individual patient samples were compared via 2-sample t tests. Univariable and multivariable logistic regression were used to evaluate factor associations with aGvHD, while Cox proportional hazard regression assessed biomarker links to NRM and OS. In multivariable logistic or Cox regression model, the combination of 4 biomarkers were simultaneously included in the model as dependent variables. Internal validation involved 1000 bootstrap resamples to quantify optimism. Diagnostic accuracy was measured by area under the receiver operating characteristic curves (AUROCs), sensitivity, specificity, positive/negative predictive values. Kaplan–Meier estimates and log-rank tests analyzed NRM/OS cumulative incidence. All tests were 2-sided, with statistical significance set at P < .05. Receiver operating characteristic (ROC) curves were compared using the DeLong method to assess differences in area under the curve (AUC), with MedCalc software (Version 23.1.1, MedCalc Software Ltd., Ostend, Belgium) employed for analysis. Other statistical analyses were all performed with Statistics 22 (IBM Corp., Armonk, NY) and Graghpad Prism8 (GraghPad Software Inc., San Diego, CA).

3. Results

3.1. Patient characteristics

Patient characteristics are shown in Table 1. No significant differences were found in age distribution, gender, diagnosis, disease status at transplant, donor-recipient human leukocyte antigen match, cluster of differentiation 34 + dose. All conditioning regimens were myeloablative and intensified as previously reported,^[19] with standardized aGvHD prevention. Of 141 patients, 47 had grade II-IV aGvHD:13 skin-aGvHD, 20 gut-aGvHD, 2 liver-aGvHD and 12 muti-organs aGvHD. The median aGvHD onset time was 32-day post allo-HSCT.

Table 1.

Patient characteristics (n = 141).

Characteristics	aGVHD N = 47	Non-GVHD N = 94	P value
Median age (range), yr	37.5 (2–67)	39 (1–68)	.152
Male, n (%)	27 (57)	53 (56)	.904
Diagnosis, n (%)			.453
Acute leukemia	43 (92)	82 (87)
Others (MDS/MPD/ NHL)	4 (8)	12 (13)
Disease Status at transplant, n (%)			.404
NR	27 (57)	47 (50)
CR	20 (43)	47 (50)
Donor-recipient HLA match, n (%)			.347
≥ 8/10	15 (32)	23 (24)
≤ 7/10	32 (68)	71 (76)
CD34 + dose, n (%)			.310
< 5 × 106/kg	18 (38)	28 (30)
≥ 5 × 106/kg	29 (62)	66 (70)
Conditioning regimen intensity, n (%)			-
High-intensity	47 (100)	94 (100)
Moderate-intensity	0	0
GvHD prophylaxis: n (%)			-
CNI + MTX + MMF	46 (98)	90 (96)
Other regimens	0	0
Involved organs, n (%)			< .001
Skin	13 (28%)	0
GI	20 (43%)	0
liver	2 (4%)	0
Multi-organs	12 (25%)	0

Abbreviations: aGvHD = acute graft-versus-host disease, ATG = antithymocyte globulin, CD34 = cluster of differentiation 34, CNI = calcineurin inhibitor, CR = complete response, GI = gastrointestinal tract, HLA = human leukocyte antigen, MDS/MPD = myelodysplastic syndrome/myeloproliferative neoplasms, MMF = mycophenolic acid, MTX = methotrexate, n/N = number of patients, NR = no response, NHL = Non-Hodgkin Lymphoma.

3.2. Biomarkers concentrations at different target organ aGvHD groups

Serum concentrations of ST2, REG3α, Elafin, and TNFR1 were compared between different target organ aGvHD (Fig. 2). ST2 concentrations were considerably higher in multi-organ GvHD than in other patients. REG3α concentrations were significantly increased in the GI-GvHD group. Elafin concentrations were higher in both multi-organ GvHD and GI-GvHD groups, and not markedly elevated in the skin-GvHD group. There was no significant difference in TNFR1 concentrations among all groups.

Figure 2.

ST2, REG3α and Elafin concentrations in different GvHD patient groups (non-GvHD, n = 94; skin-GvHD, n = 13; GI-GvHD, n = 20; and muti-GvHD, n = 12). (A) ST2 serum concentration was significantly high in muti-GvHD group patients. (B) REG3α serum concentration was notably elevated in GI-GvHD group patients. (C) Elafin serum concentration was remarkably high in muti-GvHD group patients. GI = gastrointestinal tract, GvHD = graft-versus-host disease, n = number of patients, ns = not significant, REG3α = regenerating islet-derived 3 α, ST2 = suppressor of tumorigenesis 2.

3.3. Algorithm development for predicting aGvHD, NRM and OS

Univariable and multivariable regression analyses evaluated the associations of ST2, REG3α, Elafin, and TNRF1 concentrations with 100-day aGvHD, 12-month NRM and OS were presented in Table 2. Univariable analysis indicated that ST2, REG3α, and Elafin were positively associated with aGvHD; Multivariable analysis identified ST2 as the strongest predictor (P < .001) for aGvHD.

Table 2.

Univariable and multivariable analysis of 4 biomarkers in predicting 100d aGvHD, 12-month NRM and OS.

100d aGvHD			12-month NRM		12-month OS
	(OR, 95% CI)	P	(HR, 95% CI)	P	(HR, 95% CI)	P
Univariable
ST2	2.21 (1.62–3.02)	< .001	1.38 (1.15–1.71)	.003	1.32 (1.11–1.57)	.002
REG3α	3.42 (1.94–6.05)	< .001	1.83 (1.20–2.79)	.005	1.41 (1.01–1.99)	.045
Elafin	3.58 (1.93–6.62)	< .001	1.62 (1.01–2.60)	.044	1.66 (1.14–2.45)	.009
TNFR1	1.55 (0.88–2.70)	.127	3.44 (1.97–6.01)	< .001	2.42 (1.57–3.74)	< .001
Multivariable
ST2	1.91 (1.32–2.75)	< .001	1.18 (0.91–1.53)	.210	1.14 (0.93–1.41)	.198
REG3α	2.19 (1.06–4.50)	.034	1.30 (0.76–2.22)	.340	0.90 (0.59–1.39)	.644
Elafin	1.31 (0.57–2.97)	.526	0.92 (0.47–1.82)	.809	1.30 (0.79–2.15)	.299
TNFR1	0.65 (0.31–1.33)	.239	2.82 (1.56–5.08)	< .001	2.15 (1.35–3.42)	< .001

aGvHD = acute graft-versus-host disease, AUC = area under the curve, NRM = non-relapse mortality, OS = overall survival, REG3α = regenerating islet-derived 3 α, ROC = receiver operating characteristic, ST2 = suppressor of tumorigenesis 2, TNFR1 = tumor necrosis factor 1.

During a median follow-up period of 15 months of 141 patients, 58 patients died, and 38 cases were classified as NRM. Univariable analysis showed that ST2, REG3α, Elafin, and TNRF1 were all positively associated with 12-month NRM and OS. Multivariable analysis demonstrated that TNFR1 had the highest prognostic value in predicting 12-month NRM and OS (P < .001). We compared the diagnostic and prognostic ability of ST2, REG3α, Elafin and TNRF1 for aGvHD using AUROCs (Fig. 3), sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio in Table S1, Supplemental Digital Content, https://links.lww.com/MD/R728. The panel of ST2, REG3α, Elafin, and TNFR1 demonstrated the best predictive performance in the diagnosis of aGvHD (AUC = 0.81) and in predicting of 12-month of NRM (AUC = 0.74) and OS (AUC = 0.71). We compared the ST2/ REG3α/TNFR1 (3-biomarker) algorithm to the 4-biomarker algorithm in predicting aGvHD and prognosis (Figure S1, Supplemental Digital Content, https://links.lww.com/MD/R729). For the diagnosis of aGvHD, prediction of 12-month NRM and OS, the AUC were 0.81, 0.73, and 0.69, respectively, no significant difference was observed between the 2 algorithms. Bootstrap validation (1000 iterations) yielded optimism-corrected AUCs of 0.80 for aGvHD diagnosis, 0.73 for 12-month NRM, and 0.71 for OS, suggesting minimal overfitting.

Figure 3.

ROC curves for different biomarkers with the prediction of aGvHD, NRM and OS. (A) Different biomarkers for prediction of aGvHD. ST2 (thin black): AUC = 0.79; REG3α (thin green): AUC = 0.75; Elafin (thin orange): AUC = 0.71; TNFR1 (thin pink): AUC = 0.59; (B)Composite of ST2 + REG3α+ Elafin + TNFR1(thick blue): AUC = 0.81, sensitivity = 72.3%, specificity = 77.7%. (C) Different biomarkers for prediction of NRM. ST2 (thin black): AUC = 0.64; REG3α (thin green): AUC = 0.63; Elafin (thin orange): AUC = 0.54; TNFR1 (thin pink): AUC = 0.73; (D)Composite of ST2 + REG3α+Elafin and TNFR1: AUC = 0.74, sensitivity = 71.1%, specificity = 68.0%. (E) Different biomarkers for prediction of OS. ST2 (thin black): AUC = 0.63; REG3α (thin green): AUC = 0.59; Elafin (thin orange): AUC = 0.60; TNFR1 (thin pink): AUC = 0.69. (F) Composite of ST2 + Elafin + REG3α+TNFR1: AUC = 0.71, sensitivity = 79.3%, specificity = 61.5%. aGvHD = acute graft-versus-host disease, NRM = non-relapse mortality, OS = overall survival, REG3α = regenerating islet-derived 3 α, ROC = receiver operating characteristic, ST2 = suppressor of tumorigenesis 2, TNFR1 = tumor necrosis factor 1.

3.4. Risk stratification

We determined thresholds for the “ST2 + REG3α+Elafin + TNFR1” combination panel to predict 12-month NRM and OS, then used these thresholds to risk-stratify 141 patients (Fig. 4). The 12-month cumulative NRM incidence was 43% in the high-risk group and 11% in the low-risk group. Correspondingly, 12-month OS rates were 51% and 82% in the high-risk and low-risk groups, respectively.

Figure 4.

12-month cumulative incidence of NRM and 12-month OS by risk classification for ST2 + REG3α+Elafin + TNFR1 biomarker algorithms. (A) 12-month cumulative incidence of NRM according to risk classification. 43% vs 11% at high risk (thick line) and low risk (thin line) patients respectively. (B) Kaplan–Meier estimates of 12-month OS according to risk classification. 51% vs 82% at high risk (thick line) and low risk (thin line) patients respectively. allo-HSCT = allogeneic hematopoietic cell transplantation, NRM = non-relapse mortality, OS = overall survival, REG3α = regenerating islet-derived 3 α, ST2 = suppressor of tumorigenesis 2, TNFR1 = tumor necrosis factor 1.

4. Discussion

As a research hotspot, several groups have validated biomarkers combination to predict aGvHD and outcomes.^[16–18,20] In this study, we have developed an algorithm using a combination panel of “ST2 + REG3α + Elafin + TNFR1” to predict aGvHD, 12-month NRM and OS.

The maximum severity of aGvHD is mainly driven by GI damage and correlates well with NRM and OS. REG3α has been reported for the diagnosis and risk stratification of GI aGvHD.^{[10,11,14–16,20]} ST2 is another established predicting biomarker for aGvHD which is a ligand for interleukin-33, a protein secreted by damaged epithelial cells.^{[12,13,20–22]} Besides, ST2 has also been reported to be associated with the risk of treatment-resistant aGvHD.^[23] Elafin has been reported as a classic skin aGVHD biomarker.^[11] Previous studies have shown that Elafin is the best discriminator for distinguishing skin aGVHD from other etiologies of rash, such as engraftment syndrome, leukemia cutis, and drug rash.^[24] TNFR1 is a classical proinflammatory cytokine that has been used in a biomarker panel for aGVHD diagnosis and prognosis.^[17] A systemic review showed single TNFR1 lacks sufficient accuracy for aGvHD prediction, but combining it with other biomarkers may enhance screening performance.^[9] In the present studies, liver aGvHD without GI involvement at the onset of disease was uncommon (n = 2; 4% of aGvHD patients in our study), we did not choose any liver aGvHD related biomarkers such as hepatocyte growth factor and cytokeratin fragment 18.^[25] Thus, we selected these 4 biomarkers to develop a highly specific, sensitive, rapid, and cost-effective algorithm for aGvHD diagnosis and prognosis.

We compared cytokine levels in aGvHD patients with different organ involvement. Results showed serum REG3α levels were significantly higher in those with gastrointestinal aGvHD, consistent with prior findings.^{[10,14,15,25]} ST2 levels were highest in multi-organ GvHD patients, mostly with GI and skin involvement. Multivariate analysis showed ST2 was more predictive of aGvHD than other single biomarkers (AUC = 0.79). As the value of Elafin in aGvHD is very controversial,^[26,27] here we compared the 3-biomarker (ST2/REG3α/TNFR1) and 4-biomarker (ST2/REG3α/Elafin/TNFR1) algorithms in their performance in predicting aGvHD and prognosis. Although the difference was not statistically significant, the ROCs of the 4-biomarker panel were all slightly higher than that of the 3-biomarker panel. Furthermore, our results showed that the level of Elafin was significantly higher in patients with multi-organ GvHD than in other groups; therefore, we included Elafin in the detection panel.

TNFR1 is a systemic inflammation biomarker. Our results showed no significant differences in TNFR1 levels between aGvHD and non-GvHD patients. However, its value in predicting 12-month NRM and OS was significant(AUROCs, 0.73 and 0.69, respectively), which was consistent with the previous study.^[28] Comparing AUROCs and performance of single/combined biomarkers, the panel showed the highest ROC for aGvHD prediction and NRM/OS prognosis, exhibiting robust predictive efficacy for both aGvHD diagnosis and prognosis.

Our study has several limitations. First, prospective multicenter validation is needed to expand sample size and confirm the algorithm’s clinical utility. Second, biomarker selection was non-comprehensive, excluding IL2Ra, IL8, transforming growth factor-beta, Tim-3, etc.^{[6,8,9,20,28]} Third, patients do not have multiple basal biomarker levels before the development of aGVHD and levels after, and measurement timing lacked precision, requiring evaluation across early time points (7, 14, 28 days post allo-HSCT). Fourth, we did not assess the biomarkers’ potential to predict aGvHD treatment response.

To date, the day 28 treatment response in aGvHD was used as a primary endpoint for longer-term outcomes such as NRM or survival.^[29] However, early response is an imperfect measure of treatment efficacy, as late response loss or treatment-related complications can occur, leading to aGvHD-related mortality. Our findings offer a rapid, effective, and cost-effective panel for predicting aGvHD and outcomes.

Author contributions

Conceptualization: Qi Hao, Xinyue Liu, Tingting Li, Wei Wei, Jiangying Gu, Jingbo Wang.

Data curation: Qi Hao, Xinyue Liu, Tingting Li, Wei Wei, Jiangying Gu, Jingbo Wang.

Formal analysis: Qi Hao, Xinyue Liu, Wei Wei.

Funding acquisition: Jingbo Wang.

Investigation: Qi Hao, Xinyue Liu, Tingting Li, Tianqi Qi, Shuqin Zhang, Xinhong Fei, Jiangying Gu.

Methodology: Qi Hao, Tianqi Qi.

Project administration: Qi Hao, Xinyue Liu, Wei Wei, Shuqin Zhang, Xinhong Fei, Jingbo Wang.

Resources: Qi Hao, Xinyue Liu, Tingting Li, Tianqi Qi, Shuqin Zhang, Xinhong Fei, Jiangying Gu.

Software: Qi Hao, Tianqi Qi.

Supervision: Qi Hao, Jingbo Wang.

Validation: Qi Hao, Xinyue Liu, Tingting Li, Wei Wei, Xinhong Fei, Jiangying Gu, Jingbo Wang.

Visualization: Qi Hao, Tianqi Qi, Shuqin Zhang, Jingbo Wang.

Writing – original draft: Qi Hao.

Writing – review & editing: Qi Hao, Xinyue Liu, Tingting Li, Wei Wei, Tianqi Qi, Shuqin Zhang, Xinhong Fei, Jiangying Gu, Jingbo Wang.