The MAGIC Marker: Measurements of Two Serum Biomarkers Predict GVHD Treatment Response

Measuring damage to crypts in the gastrointestinal tract during graft-versus-host disease (GVHD) via the Mount Sinai Acute GVHD International Consortium (MAGIC) algorithm probability (MAP) tool successfully predicted patients’ long-term outcomes, including 6-month non-relapse mortality (NRM). MAP, which is computed from levels of two serum biomarkers, also more accurately predicted durability of response to anti-GVHD therapy compared with changes in clinical symptoms.

These findings were published in Blood Advances and presented at the 2019 American Society of Hematology Annual Meeting. The study’s first author, Hrishikesh K. Srinagesh, from the Icahn School of Medicine at Mount Sinai, received the Outstanding Abstract Achievement Award for a medical student at the meeting.

“The exciting thing about the MAP is that it acts as a combination of two biomarkers that are put together into an equation,” explained corresponding author James L. M. Ferrara, MD, DSc, Professor and Director of the Hematologic Malignancies Translational Research Center at the Icahn School of Medicine at Mount Sinai. “These two biomarkers in the equation act as a ‘liquid biopsy’ that tells you about the extent of damage to the crypts, and it’s the crypt damage that really determines whether the gastrointestinal tract can heal.”

To validate the MAP’s role as a response biomarker, researchers computed MAPs for 615 patients who underwent hematopoietic cell transplantation at 20 MAGIC centers across the globe. MAP was calculated as levels of two serum biomarkers: regenerating islet-derived 3α (REG3α) and suppressor of tumorigenesis 2 (ST2). Participants were enrolled at the time of transplant; serum samples were prospectively collected at initiation of firstline systemic treatment and 4 weeks later.

Patients were divided into training (n=248) and validation (n=367) cohorts. The validation cohort was used to test the original MAP model plus multivariable models that combined the MAP with clinical responses to treatment. For 6 months, investigators monitored patients for signs and symptoms of acute GVHD.

Nonresponse to treatment was applied to patients who met the following patient criteria:

  • GVHD symptoms progressed or did not improve
  • Additional systemic immunosuppression treatment for GVHD had to be prescribed
  • Death within first 4 weeks of therapy

Patients who were considered responders had either complete resolution of disease symptoms in all three target organs or improvement in stage of GVHD involvement in all organs but without complete resolution of disease symptoms, which might have been due to causes other than GVHD.

A multivariable model incorporating both clinical responses and MAPs better predicted likelihood of sustained remission compared with clinical responses alone. MAPs also predicted NRM better than change in clinical symptoms in all patients and also identified 2 groups with significantly different rates of NRM, regardless of response. Among clinical responders, 10% had high MAPs (10%); these patients had significantly greater 12-month NRM compared with patients in the low-MAP group (40% vs. 12%, respectively; p<0.0001).

“Because the MAP is both more specific and more sensitive than the change in clinical symptoms, it gives a more accurate picture of the extent of GVHD.”

—James L. M. Ferrara, MD, DSc

To test whether the combination of clinical responses and MAPs would improve the prediction of long-term outcomes after 4 weeks of treatment, compared with either metric alone, the investigators then created a new model that combined 4-week clinical response with concentrations of REG3α and ST2. At the beginning of treatment, patients with a low MAP that rose above the threshold of 0.290 after 4 weeks of treatment had a significant increase in NRM. However, patients with a high MAP at onset that eventually fell below that threshold had “a striking decrease in NRM that translated into clear differences in overall survival.”

In terms of future applicability, Dr. Ferrara sees the MAP being used at the start of treatment as well as 1 or 2 weeks into treatment to see how well the therapy is working. “The MAP will be particularly helpful in patients who have not responded clinically or who seem slow to respond to therapy,” he said. “Because the MAP is both more specific and more sensitive than the change in clinical symptoms, it gives a more accurate picture of the extent of GVHD, particularly in the gastrointestinal tract.”

Dr. Ferrara added that, according to the results of this validation study, the MAP gives a more accurate prediction of long-term outcome at 1 week than the clinical response to treatment after 4 weeks. “The treating clinician is getting more accurate information earlier and can therefore adjust treatment accordingly.”

Currently, the MAP is being used as an eligibility criterion in GVHD therapy clinical trials at Mt. Sinai. Dr. Ferrara believes that in the next few years, researchers will begin testing it as an endpoint, either alone or together with clinical response. The ultimate goal, he said, is to use the MAP to assist in the development of new treatments for acute GVHD.

Dr. Ferrara and two co-authors report holding a GVHD biomarker patent; the remaining authors report no conflicts of interest.


Srinagesh HK, Özbek U, Kapoor U, et al. The MAGIC algorithm probability is a validated response biomarker of treatment of acute graft-versus-host disease. Blood Adv. 2019;3:4034-4042.