Autologous bone marrow mononuclear cells to treat severe traumatic brain injury in children

Abstract

Autologous bone marrow mononuclear cells (BMMNCs) infused after severe traumatic brain injury have shown promise for treating the injury. We evaluated their impact in children, particularly their hypothesized ability to preserve the blood–brain barrier and diminish neuroinflammation, leading to structural CNS preservation with improved outcomes.

We performed a randomized, double-blind, placebo-sham-controlled Bayesian dose-escalation clinical trial at two children's hospitals in Houston, TX and Phoenix, AZ, USA (NCT01851083). Patients 5–17 years of age with severe traumatic brain injury (Glasgow Coma Scale score ≤ 8) were randomized to BMMNC or placebo (3:2). Bone marrow harvest, cell isolation and infusion were completed by 48 h post-injury. A Bayesian continuous reassessment method was used with cohorts of size 3 in the BMMNC group to choose the safest between two doses. Primary end points were quantitative brain volumes using MRI and microstructural integrity of the corpus callosum (diffusivity and oedema measurements) at 6 months and 12 months. Long-term functional outcomes and ventilator days, intracranial pressure monitoring days, intensive care unit days and therapeutic intensity measures were compared between groups.

Forty-seven patients were randomized, with 37 completing 1-year follow-up (23 BMMNC, 14 placebo). BMMNC treatment was associated with an almost 3-day (23%) reduction in ventilator days, 1-day (16%) reduction in intracranial pressure monitoring days and 3-day (14%) reduction in intensive care unit (ICU) days. White matter volume at 1 year in the BMMNC group was significantly preserved compared to placebo [decrease of 19 891 versus 40 491, respectively; mean difference of −20 600, 95% confidence interval (CI): −35 868 to −5332; P = 0.01], and the number of corpus callosum streamlines was reduced more in placebo than BMMNC, supporting evidence of preserved corpus callosum connectivity in the treated groups (−431 streamlines placebo versus −37 streamlines BMMNC; mean difference of −394, 95% CI: −803 to 15; P = 0.055), but this did not reach statistical significance due to high variability.

We conclude that autologous BMMNC infusion in children within 48 h after severe traumatic brain injury is safe and feasible. Our data show that BMMNC infusion led to: (i) shorter intensive care duration and decreased ICU intensity; (ii) white matter structural preservation; and (iii) enhanced corpus callosum connectivity and improved microstructural metrics.

In a randomized controlled trial, Cox Jr et al. find that autologous bone marrow mononuclear cell infusion in children within 48 h after severe traumatic brain injury is safe and feasible and results in decreased intensive care duration and intensity, as well as better preservation of brain structure and functionality.

Introduction

Paediatric traumatic brain injury is a leading cause of death and disability in children. More than 1 million children suffer traumatic brain injuries each year, with 30 000 of those resulting in lifelong disabilities.^1,2 Currently, no effective therapies are available to treat secondary brain injury and post-injury CNS apoptosis and neuroinflammation. However, treating acute neurological injuries with various progenitor cell types has shown increasing promise, with efficacy in preclinical models of traumatic brain injury/stroke.³ The most likely mechanism of action is modification of the regional injury response by the infused cells, resulting in control of the inflammatory response and improved functional outcomes.⁴

Mononuclear cells derived from autologous bone marrow (bone marrow mononuclear cells, BMMNCs) have numerous advantages for the potential treatment of severe traumatic brain injury. Preclinical proof-of-concept studies in traumatic brain injury,^4-10 stroke,^11-13 spinal cord injury^14-18 and other fields^19-25 support safe and effective use of BMMNCs, with no serious adverse events or toxicity demonstrated in phase 1/2a trials for traumatic brain injury and stroke.^6,26,27 Additionally, autologous BMMNCs have no immune barrier considerations, and a biologically sound route of delivery is possible due to the 5–8 μm cell size, making intravenous delivery practical.²⁸ Autologous BMMNCs are also readily available; no in vitro culture or scaling is required. Further, BMMNCs have no issues with uncontrolled replication, as with embryonic stem cells, and no ethically controversial source material is required. Lastly, data already exist detailing the mechanism of action.^4,29

MRI can be used to assess structural integrity/preservation of brain tissue, which has been correlated with improved functional outcomes. Previous volumetric studies in severe adult and paediatric traumatic brain injury patients have found patients with loss of total intracranial solid brain volume (ICV), grey matter (GM) volume and white matter (WM) volume and concomitant increases in CSF space when compared to controls at 1-year post-injury.^30-33 In addition to generalized volume loss, selective loss of volume and microstructural integrity in the corpus callosum (CC) have been found in paediatric traumatic brain injury patients; these findings have been correlated anatomically to neurocognitive outcomes.^34-36 Consequently, the use of volumetric measurements as an imaging biomarker is a valid, reproducible end point that reflects the biological impact of neuroinflammatory-induced cerebral atrophy. The scientific premise of this has been confirmed in numerous clinical studies in adults and children and validated in preclinical models.³⁷ Furthermore, obtaining microstructural metrics is non-invasive, and these metrics represent potential predictive readouts linked to the evolution of cerebral oedema.

Data from our phase 1 study of paediatric traumatic brain injury patients suggests treatment with BMMNCs not only prevents post-injury GM or WM volumetric loss but that these findings persist 6 months after injury.⁶ This phase 2 study aimed to extend our initial results by determining if intravenous infusion of BMMNCs after severe traumatic brain injury in children results in structural preservation of global GM and WM and regions in the CC known to be correlated with neurocognitive function. We sought to evaluate the volumetric and microstructural changes over time at the immediate post-injury period and 1 month, 6 months and 1 year post-injury. While our primary end point was at 6 months, we studied longer-term changes due to the known progression of volumetric changes over time.

Patients and methods

Study design

This study was a phase 2 study analysing both safety and efficacy with a two-centre, randomized, double-masked, placebo-controlled, Bayesian adaptive dose-escalation design. Ethics approval was obtained from The University of Texas Health Science Center at Houston Committee for the Protection of Human Subjects (HSC-MS-13-0038) and the Memorial Hermann Hospital System, as well as the Phoenix Children's Hospital Institutional Review Board (IRB-15-136).

Data and safety monitoring and trial registry

A Data and Safety Monitoring Board (DSMB) was used for this trial. A medical safety monitor (MSM) masked to treatment reviewed serious adverse events following post-infusion Day 14 for each participant and remained masked during the trial unless the DSMB approved unmasking. Independent medical monitoring was provided by Juno Research, who worked with the MSM and DSMB to ensure safety of trial participants. This study was conducted under Federal Drug Administration Investigational New Drug Application CBER BB12620 and registered with ClinicalTrials.gov (NCT01851083).

Participants

Participants were recruited from children with acute traumatic brain injuries from closed head trauma admitted to the paediatric intensive care unit (ICU) (<16 years) at Children's Memorial Hermann or the shock trauma ICU or neurotrauma ICU at Memorial Hermann-Texas Medical Center (16–17 years) in Houston, TX or the paediatric ICU at Phoenix Children's Hospital in Phoenix, AZ; both sites are American College of Surgeons-verified level 1 paediatric and adult trauma centres. Written consent was obtained from the child's parent(s) or legally authorized representative(s) after the nature and possible consequences of the study were explained according to the Declaration of Helsinki.

A total of 50 participants were planned for enrollment. Inclusion criteria included age of 5–17 years on day of injury and Glasgow Coma Scale score (GCS) 3–8.³⁸ Exclusion criteria included known previous brain injury, obliteration of the perimesencephalic cistern, initial hospital intracranial pressure (ICP) > 40 mm Hg and penetrating brain injury. Supplementary Table 1 details full eligibility criteria. The site principal investigator made final determination on participant eligibility before randomization.

Randomization and masking

Participants were randomized in a 3:2 ratio of autologous BMMNC to placebo infusion as a single dose given within 48 h after injury. Randomization was done using permuted varying block sizes and stratified by GCS score at baseline (3–4 or 5–8). Patients randomized to BMMNC were allocated to receive a low dose of 6 × 10⁶ cells/kg body weight or a high dose of 10 × 10⁶ cells/kg body weight using a Bayesian adaptive dose-escalation method. Randomization lists were prepared by the study statistician. The participant's family, neurocognitive assessors, imaging team and clinical care providers were masked to randomization assignment. The procedures to ensure masking for infusions are described below.

Bayesian continual reassessment method

A Bayesian continual reassessment method (CRM) model was used to guide dose escalation and de-escalation and help assure safe dosing levels. The Bayesian CRM models the relationship between dose and the risk of a dose-limiting toxicity (DLT). Based on the probability of a DLT (defined below), the algorithm indicated whether to escalate or de-escalate the dose. Initially, the DLT definition was meant to only include infusion-related toxicities, and hence, the target toxicity level was initially set at 15%. However, after the fourth cohort, we identified that clinicians were recording all toxicities as DLTs, and we, therefore, increased the target toxicity rate to 30%. A 1-parameter logistic CRM with the intercept term fixed at 1 and a Normal prior with mean 0 and variance 1.8 for the slope parameter was used. The starting dose was the low dose, and the algorithm updated the probabilities of DLT for both doses after each cohort of three children treated with autologous BMMNC infusion. The recommended dose level for the next cohort was the dose with updated DLT rate closest to the target level of 30%, incorporating DLT outcomes for all treated children up to that point. The algorithm indicated whether to stay at the same dose, escalate or de-escalate the dose or stop the trial (if the probability of the lowest dose having a DLT rate greater than 30% exceeded 80%). Separate CRMs were run for the two GCS strata (3–4, 5–8). See the Supplementary material ‘Methods’ section for more information.

Traumatic brain injury management

Established traumatic brain injury management guidelines were followed after injury (Supplementary material ‘Methods’ section).

Bone marrow harvest, sham harvest procedure and bone marrow processing and infusion

Bone marrow harvesting, sham harvesting and bone marrow and placebo processing were carried out as described by Cox et al.⁵ (Supplementary material ‘Methods’ section and Supplementary Fig. 1).

Follow-up

Follow-up assessments were completed at 1 month (±15 days), 6 months (±21 days) and 1 year (±21 days) post-injury.

Outcomes

The primary outcome measures were to: (i) determine the effect of autologous BMMNC infusion on diffusion tensor (DT)-MRI-determined CNS WM, GM and CC volumetric preservation; and (ii) determine if autologous BMMNC infusion preserves microstructural integrity of GM and WM brain regions of interest (principally the CC) as measured by DT-MRI-determined metrics of mean (MD), axial (AD) and radial diffusivity (RD) and fractional anisotropy (FA). These structural outcomes were examined along with functional and neurocognitive deficits in children after traumatic brain injury. BMMNC infusion safety was also evaluated.

Safety monitoring for infusional toxicity and organ function

DLT was defined as (i) 30% drop in the PaO₂:FiO₂ ratio from baseline within first 72 h post-infusion; OR (ii) absolute value of <250 within first 72 h post-infusion and not associated with nosocomial pneumonia; OR (iii) Murray score³⁹ of over 2.5 when all components of score were obtained during the same time window. The Pediatric Logistic Organ Dysfunction (PELOD) score, a measure of injury severity in paediatric patients,^40,41 was calculated daily. If the patient was sedated with concomitant neuromuscular blockade, the pre-sedation GCS was used to calculate the PELOD score. The Murray score, or lung injury score, is a measure of acute lung injury³⁹ and was calculated daily. The Pediatric Intensity Level of Therapy (PILOT) scale is a standardized measure used to compare overall therapeutic intensity required to manage ICP in severe paediatric traumatic brain injury.^7,42,43 The score is calculated every 24 h following initial trauma. See the Supplementary material ‘Methods’ section for further description.

Additional safety monitoring during bone marrow/sham harvest, including vital signs and ICP, was performed. Post-infusion monitoring continued per ICU standard of care. For participants receiving BMMNC infusion, a toxicity assessment was done 72 h after infusion.

Functional and neuropsychological outcome measures

Glasgow Outcome Score-Extended Peds (GOS-E Peds),⁴⁴ Disability Rating Scale for Children (DRS-C; L. Ewing-Cobbs and D. R. Bloom; Disability rating scale for children after brain injury; unpublished work) and Pediatric Injury Functional Outcome Scale (PIFOS)⁴⁵ scores were assessed via a structured interview at each follow-up. The interviewers were trained by the Neuropsychology group (Ewing-Cobbs) and masked to treatment group. The primary caregiver completed standardized neuropsychology questionnaires, including the Adaptive Behavior Assessment System II (ABAS),⁴⁶ to indicate performance just prior to injury and at 6- and 12-month follow-up visits. Supplementary Table 2 provides descriptive information and lists dependent variables of interest for each measure.

Multimodal MRI

The planned MRI acquisition schedule included four time points: acute post-injury, 1 month post-injury, 6 months post-injury and 1 year post-injury. The acute time point was the earliest window after ICP monitor removal (usually about 5–7 days after injury) and was done while the participant was still mechanically ventilated and did not require additional sedation. The subsequent time points were done within a window with potential minor changes based on medical factors. Details about MRI acquisition at each study centre (i.e. study instruments, MRI sequences and imaging time) are in the Supplementary material ‘Methods’ section.

Statistical analysis

Sample size

A sample size of 50 participants (30 BMMNC and 20 placebo) was calculated based on a hypothesized mean difference of 87 cm³ in total whole brain GM at 6 months (mean volume of 884 cm³ in BMMNC group versus 797 cm³ in placebo group; standard deviation of 77 cm³). This volume represents a 10% volumetric loss, which is consistent with the literature that is associated with severe traumatic brain injury and the resultant cognitive deficits.^47-49 Also incorporated into the sample size calculation was a 20% loss to follow-up rate, two-sided alpha of 0.05 and 90% power.

Analyses

Primary analyses were based on principles of intention to treat including all participants with at least one post-randomization outcome at 6 months or 1 year. High- and low-dose BMMNC groups were analysed as one group per protocol. Volumetric/DT-MRI and neurocognitive outcomes were analysed with linear mixed models, including time (baseline, 1 month, 6 months, 1 year), group (placebo/BMMNC), time-group interaction, GCS score (3–4 or 5–8; stratifying variable) and centre as covariates. We evaluated a quadratic term for time for each outcome. A random intercept was included in all models to account for within-patient correlation. GOS-E Peds scores were dichotomized and analysed with logistic mixed models, including the same covariates and random effects. Daily PILOT, PELOD and ICP scores were analysed by calculating the area under the curve (AUC) for each participant using the linear trapezoidal method from day of infusion to the last day with a recorded score. The AUCs were compared between groups with a linear regression model including group, GCS score and centre as covariates. Numbers of ICU, ventilator and ICP days were analysed with negative binomial regression models including the same covariates. We report group mean differences, odds ratios (OR) or risk ratios (RR) for all models and 95% confidence intervals (CIs). Linear mixed models assume that data are missing at random, and we evaluated whether any baseline characteristics were associated with missingness of the primary outcome to test the assumption. No adjustment was made for multiple testing. All analyses were conducted using R software version 4.0.3.⁵⁰

One unplanned interim futility analysis was conducted at the request of the DSMB when 33 participants had been enrolled (see the Supplementary material ‘Methods’ section). The DSMB recommended that the trial continue as planned to complete enrollment to sample size.

MRI analyses

Amplified MRI analyses

The high-resolution amplified MRI sequences were used in the longitudinal Freesurfer v7.1.0 image processing stream⁵¹ to automatically segment and parcellate the brain into labeled regions of interest, including the midsagittal CC. While the 3D T1-weighted sequence was the primary sequence used by Freesurfer, additional refinement of the pial surface reconstruction was achieved using the 3D T2-weighted fluid-attenuated inversion recovery (FLAIR).

Diffusion MRI analyses

The diffusion MRI (dMRI) sequences differed slightly between the two study sites due to vendor-specific platform differences. However, longitudinal time points within participant were rigorously acquired on the same MRI scanner at each site. FSLv6.0.3⁵² was used for preprocessing dMRI data, scalar map reconstruction and co-registering structural MRI (sMRI) with dMRI for subsequent analyses. Preprocessing steps included FSL's top-up and eddy functions to correct for geometric and motion-related distortions in the dMRI data. Subsequently, FSL's tensor fitting function (DTIfit) was used to derive common scalar maps, including MD. Additionally, FSL's FLIRT function was used to co-register (within participant) the sMRI series to each other and to the dMRI series based on the T2-weighted and b = 0 s/mm² volumes. A quick three-class tissue segmentation (GM, WM, CSF) was performed using FSL's FAST function on the T1-weighted series, which was already co-registered (transformed) to the native space of the T2-weighted series. MD scalar maps were used to extract MD values globally from each tissue class and regionally from the midsagittal CC. Finally, deterministic tractography methods (TrackVis⁵³) were used to reconstruct the connectivity of the CC with the rest of the brain at each time point. The number of streamlines was documented quantitatively and recorded at each time point. All personnel involved in image analysis were masked to treatment.

Results

This two-centre, randomized, double-masked, placebo-sham-controlled trial was conducted between 25 August 2013 and 16 September 2020. Figure 1 details participant enrollment and randomization, as well as assessment for eligibility and dropouts. Of note, enrollment was halted at 47 participants due to wind-up of funding. However, we were able to assess the primary outcome effectively and in a statistically meaningful way with this number of patients.

Figure 1.

Patient flow diagram. BMMNC = bone marrow mononuclear cells; PCH = Phoenix Children’s Hospital; UTH = UTHealth Houston.

Table 1 details baseline patient characteristics. Supplementary Table 3 presents these characteristics and compares participants with/without 6- or 12-month imaging outcomes observed. We note no important differences between the two groups.

Table 1.

Baseline characteristics

	BMMNC	Placebo
Age, years, median (IQR)	14.6 (10.2, 15.7)	15.4 (11.4, 16.4)
Sex, n (%)
Male	17 (63.0)	14 (73.7)
Female	10 (37.0)	5 (26.3)
Race/Ethnicity, n (%)
American Indian or Alaskan Native	0 (0.0)	1 (5.3)
Asian	2 (7.4)	0 (0.0)
Black or African American	8 (29.6)	3 (15.8)
Hispanic	9 (33.3)	10 (52.6)
White	8 (29.6)	5 (26.3)
GCS at randomization, median (IQR)	7.0 (6.0, 7.0)	7.0 (6.0, 7.0)
GCS group, n (%)
3–4	2 (7.4)	2 (10.5)
5–8	25 (92.6)	17 (89.5)
PILOT scores at admission
n	22	15
Mean (SD)	10.1 (4.9)	10.2 (6.3)
PELOD scores at admission
n	17	10
Mean (SD)	10.3 (11.8)	8.6 (6.3)
Murray lung injury scores at admission
n	16	10
Mean (SD)	0.8 (0.8)	0.5 (0.6)

BMMNC = bone marrow mononuclear cells; GCS = Glasgow coma scale score; IQR = interquartile range; PELOD = Pediatric Logistic Organ Dysfunction; PILOT = Pediatric Level of Intensity Therapy; SD = standard deviation.

Bayesian CRM

The assigned BMMNC dose levels and observed DLTs of the 27 treated patients are listed in Supplementary Table 4. DLTs were recorded in three of nine participants who received the low dose and in 8 of 18 participants who received the high dose. Based on the eight completed participant cohorts with baseline GCS of 5–8, the low dose of BMMNC had an estimated posterior probability of DLT of 36% (95% credible interval: 18%–56%). We had some concerns that the higher dosing target, being based on volume of bone marrow, would be hard to consistently reach. This turned out to be true (Supplementary Table 5). This is one of the reasons we chose a priori to look at dosing as a pooled data readout.

Safety

All adverse events are detailed in Supplementary Table 6. There were no episodes of haemodynamic changes of significance, pathological elevations of ICP, harvest site haematoma or infection due to bone marrow harvest. There were no episodes of infusion-related toxicity/organ injury. Specifically, there were no transfusion reactions or anaphylactic reactions (urticarial rash, bronchial reactivity, hypotension, haemolysis). One patient in the BMMNC group died before the 1-month follow-up visit due to sepsis.

Pediatric Intensity Level of Therapy and Pediatric Logistic Organ Dysfunction scores

PILOT^42,43 and PELOD^40,41 scores as measures of organ injury and therapeutic intensity to manage ICP were assessed.

Pediatric Intensity Level of Therapy

At admission, mean PILOT scores were 10.2 (6.2) for placebo and 10.1 (4.9) for BMMNC groups. On the day of infusion, the mean PILOT scores were 8.3 (4.6) for placebo and 7.3 (4.0) for BMMNC groups (Table 2). PILOT scores decreased after infusion across time for both groups. When analysing these scores with a linear mixed model accounting for longitudinal aspects, there appeared to be a linear trend, with the scores decreasing by 8.5 (95% CI, 6.7 to 10.2) points each day, which did not differ significantly between groups (P = 0.21). After calculating AUC for each patient and comparing group means, we found mean AUC did not differ significantly between groups (mean difference −4.9; 95% CI, −21.4 to 11.6; P = 0.55).

Table 2.

Pediatric Intensity Level Of Therapy (PILOT) scores, Pediatric Logistic Organ Dysfunction (PELOD) scores, Murray lung injury scores

	PILOT		PELOD		Murray
	BMMNC	Placebo	BMMNC	Placebo	BMMNC	Placebo
Admission
n	22	15	17	10	16	10
Mean (SD)	10.1 (4.9)	10.2 (6.3)	10.3 (11.8)	8.6 (6.3)	0.8 (0.8)	0.5 (0.6)
Harvest/Infusion day
n	24	18	26	19	26	18
Mean (SD)	7.3 (4.0)	8.3 (4.6)	10.8 (8.8)	11.2 (8.5)	0.9 (0.7)	0.8 (0.7)
Post-harvest Day 1
n	25	18	27	19	27	18
Mean (SD)	5.3 (2.1)	6.2 (3.5)	9.4 (8.2)	11.7 (10.6)	1.2 (1.0)	1.0 (0.8)
Post-harvest Day 2
n	22	17	27	19	24	18
Mean (SD)	5.6 (2.9)	6.9 (3.3)	8.3 (9.0)	8.1 (8.8)	1.3 (1.0)	1.1 (0.9)
Post-harvest Day 3
n	17	15	26	19	23	18
Mean (SD)	6.1 (2.7)	5.9 (3.1)	6.9 (8.4)	5.9 (7.6)	1.1 (0.9)	1.1 (0.9)
Post-harvest Day 4
n	15	14	25	18	18	15
Mean (SD)	5.7 (2.4)	5.1 (3.4)	7.4 (8.1)	6.6 (7.1)	1.3 (0.9)	1.1 (0.7)
Post-harvest Day 5
n	14	11	23	18	18	14
Mean (SD)	6.0 (2.7)	5.9 (2.8)	5.5 (8.5)	7.4 (8.8)	1.2 (0.8)	1.1 (0.8)
Estimated treatment contrast (95% CI)	1.7 (−2.2, 5.7)a		0.13 (−0.14, 0.4)b		−0.01 (−0.06, 0.04)a
P-value	0.21		0.34		0.59

BMMNC = bone marrow mononuclear cells; CI = confidence interval; SD = standard deviation.

^aEstimated difference between groups in slope obtained from linear mixed model with an interaction between time and treatment group.

^bEstimated difference between groups in linear term obtained from linear mixed model with an interaction between time and treatment group.

Secondary analyses including the PILOT scores from day of admission were also performed and yielded similar results. Scores decreased 9.8 (95% CI, 7.7 to 11.7) points each day, with no evidence of group differences (P = 0.40); mean AUC group difference −5.9 (95% CI, −23.0 to 11.3; P = 0.49).

Pediatric Logistic Organ Dysfunction

At admission, mean PELOD scores were 8.6 (6.3) for placebo and 10.3 (11.8) for BMMNC groups; mean scores on day of infusion were 11.2 (8.5) for placebo and 10.8 (8.8) for BMMNC groups (Table 2). PELOD scores decreased across time similarly in both groups and declined at a greater rate in the BMMNC group; however, there was no statistical difference between groups. When analysing these scores with a linear mixed model accounting for the longitudinal aspect, there appears a curvilinear trend with the scores decreasing from Day 1 to Day 5–6 and then leveling after that. The curvilinear trend did not differ significantly between groups (quadratic difference −0.02; 95% CI, −0.11 to 0.06; P = 0.75). After calculating AUC for each patient and comparing group means, we found mean AUC did not differ significantly between groups (mean group difference −15.1; 95% CI, −53.1 to 22.8; P = 0.42).

Murray score for acute lung injury

There were no differences between groups as assessed by Murray scores³⁹ (Table 2).

Intensive care unit utilization

We quantified ICU utilization (ventilator days, ICP monitoring days, ICU days) as a surrogate measure of clinical efficacy. The BMMNC group trended toward lower number of ventilator days, ICP monitoring days and ICU days with RR of 0.77–0.86, but with wide 95% CIs (Table 3).

Table 3.

Intensive care unit utilization

	Group
	BMMNC (n = 27)	Placebo (n = 19)	Risk ratioa (95% CI)
Intensive care unit days
Mean (SD)	11.5 (5.9)	14.2 (10.6)	0.86 (0.61, 1.2)
Median (IQR)	11.0 (6.5, 14.5)	12.0 (7.0, 16.0)	–
Ventilator days
Mean (SD)	8.1 (4.7)	10.6 (8.7)	0.77 (0.52, 1.14)
Median (IQR)	7.0 (4.0, 12.0)	7.0 (4.5, 14.0)	–
Intracranial pressure monitoring days
Mean (SD)	5.7 (3.6)	6.8 (5.1)	0.84 (0.54, 1.3)
Median (IQR)	6.0 (3.0, 8.0)	6.0 (3.5, 9.5)	–

BMMNC = bone marrow mononuclear cells; CI = confidence interval; IQR = interquartile range; SD = standard deviation.

^aReference for the risk ratio is placebo group, and the risk ratio is adjusted for stratification variable of Glasgow coma scale score at randomization (3–4 versus 5–8) and study centre.

Longitudinal functional and neuropsychological outcome measures

Table 4 and Fig. 2 show longitudinal functional outcome scores by group. For GOS-E Peds, ORs were estimated examining effects of group and time, with scores dichotomized into good/poor outcomes.

Table 4.

Longitudinal functional outcome measures by group

Time point	1 month		6 months		1 year
Group	BMMNC	Placebo	BMMNC	Placebo	BMMNC	Placebo
n	27	17	23	17	24	14
Disability Rating Scale for Children, mean (SD)	14.7 (7.8)	14.9 (8.9)	7.8 (6.7)	9.3 (8.4)	6.1 (7.1)	7.9 (7.1)
Pediatric Injury Functional Outcome Scale, mean (SD)	67.9 (24.5)	67.8 (25.6)	52.7 (17.8)	54.2 (25.8)	46.6 (19.8)	47.2 (21.2)

BMMNC = bone marrow mononuclear cells; SD = standard deviation.

Figure 2.

Glasgow Outcome Score-Extended Peds (GOS-E Peds) scores. Placebo and BMMNC groups at 1 month, 6 months and 1 year. Percentage of participants with good scores (1–4) and bad scores (5–8) at 1 month, 6 months and 1 year after infusion are indicated. GOSE-E Peds scale: 1 = upper good recovery; 2 = lower good recovery; 3 = upper moderate disability; 4 = lower moderate disability; 5 = upper severe disability; 6 = lower severe disability; 7 = vegetative; 8 = dead. BMMNC = bone marrow mononuclear cells. Created with Datawrapper.

Glasgow Outcome Score-Extended Peds

Scores were dichotomized, with poor being 5 to 8 (or vegetative to upper severe disability categories). Patients in both groups functioned at the lower to upper severe disability level at 1-month [82% (14/17) in the placebo group and 89% (24/27) in the BMMNC group] and 6-month follow-ups [59% (10/17) in the placebo group and 43% (10/23) in the BMMNC group], with improvement to lower to upper moderate disability by 1 year [43% (6/14) in the placebo group and 29% (7/24) in the BMMNC group] (Fig. 2). Stated differently, at 1 year, the odds of a good outcome with treatment were 71% compared to 57% in placebo. In both groups, odds of bad outcome decreased significantly at 6 months (OR, 0.086; 95% CI, 0.02 to 0.34) and 1 year (OR, 0.044; 95% CI, 0.01 to 0.18) compared to 1 month. No difference was found between groups, meaning the main effect of group and group × time interaction were not significant.

Neuropsychological measures

General linear mixed model analyses of DRS-C and PIFOS scores also documented significant improvement over time, indicated by lower scores, regardless of group. DRS-C scores declined significantly at 6 months (mean decrease 6.4 points; 95% CI, 4.4 to 8.4; P < 0.001) and 1 year (mean decrease 8 points; 95% CI, 6 to 10; P < 0.001); PIFOS scores declined at 6 months (14.5 points; 95% CI, 8.6 to 20.5; P < 0.001) and 1 year (20.2 points; 95% CI, 4.4 to 26; P < 0.001) (Table 4). The ABAS scores demonstrated no differences between groups in any of the tested domains, and the group × time interaction was not significant (Supplementary Table 7).

MRI longitudinal structural findings

Estimated differences across time and group for the longitudinal WM, GM and total brain volumes (not including ventricle), CC volumes and microstructural indices are shown in Table 5. WM volume at 1 year in the BMMNC group was significantly preserved compared to placebo. Although there was greater volumetric loss in the placebo group, there was no significant difference between groups in GM or total volume. There was also no significant difference in CC volumes between groups. However, the microstructural index measurement of AD was higher in the placebo group as compared to the BMMNC group at 6 months and 1 year; the MD also trended higher in the placebo group compared to the BMMNC group. There were no differences in RD or FA between groups. Additionally, based on deterministic tractography of the CC connectivity over time, we found the number of CC streamlines was reduced more in placebo than treated participants, supporting evidence of preserved CC connectivity in the treated group (−431 streamlines placebo versus −37 streamlines BMMNC; group mean difference of −394; 95% CI, −803 to 15; P = 0.055) at 1-year follow-up, but this treatment and 11-fold decrease did not achieve statistical significance due to high variability, despite a large effect size of 0.83 SD.

Table 5.

Longitudinal structural findings

Global structural MRI metrics	Deltaa,b placebo	Deltaa,b BMMNC	Delta group differenceb,c (95% CI)	P-value
Six-month change
n	19	26	–	–
Cerebral WM volume, mm3	−15 609	−10 280	−5329 (−14430, 3772)	0.24
Supra-tentorial volume, mm3 (not ventricular)	−31 403	−54 207	22 805 (−20924, 66533)	0.29
Total GM volume, mm3	−19 962	−24 916	4955 (−33926, 43835)	0.80
n	19	25	–	–
Whole corpus callosal volume, mm3	−196	−240	44 (−268, 355)	0.78
Fractional anisotropy CC	−0.023	−0.03	0.006 (−0.018, 0.031)	0.60
Mean diffusivity CC, mm2/s	0.199	0.129	0.069 (−0.014, 0.153)	0.09e
Axial diffusivity CC, mm2/s	0.27	0.14	0.13 (0.008, 0.256)	0.03d
Radial diffusivity CC, mm2/s	0.16	0.13	0.03 (−0.048, 0.115)	0.41
Midsagittal CC connectivity
CC streamlines (number)	−308	−5.3	−303 (−791, 185)	0.21
CC mean length streamlines, mm	1.6	−0.4	1.9 (−5.9, 9.7)	0.62
One-year change
n	19	26	–	–
Cerebral WM volume, mm3	−40 491	−19 891	−20 600 (−35868, −5332)	0.01d
Supra-tentorial volume, mm3 (not ventricular)	−53 522	−33 268	−20 254 (−57679, 17171)	0.28
Total GM volume, mm3	−28 230	−9627	−18 603 (−50628, 13422)	0.25
n	19	25	–	–
Whole corpus callosal volume, mm3	−241	−186	−55 (−339, 229)	0.70
Fractional anisotropy CC	−0.02	−0.03	0.0098 (−0.01, 0.03)	0.39
Mean diffusivity CC, mm2/s	0.23	0.17	0.0653 (−0.01, 0.15)	0.09e
Axial diffusivity CC, mm2/s	0.32	0.20	0.1220 (0.008, 0.24)	0.03d
Radial diffusivity CC, mm2/s	0.18	0.15	0.0302 (−0.045, 0.11)	0.42
Midsagittal CC connectivity
CC streamlines (number)	−431	−37	−394 (−803, 15)	0.055e
CC mean length streamlines, mm	−2.6	0.5	−3.1 (−10.4, 4.1)	0.39

BMMNC = bone marrow mononuclear cells; CC = corpus callosum; GCS = Glasgow coma scale; GM = grey matter; WM = white matter.

^aDelta = estimates of difference from baseline to 6-month or 1-year time point (negative values indicate decreases).

^bEstimates obtained from linear mixed model including time (baseline, 1 month, 6 months, 1 year), group (placebo/BMMNC), time-group interaction, GCS score (3–4 or 5–8) and centre as covariates and a random intercept for participant.

^cDelta group difference = Delta placebo − Delta BMMNC.

^dStatistically significant differences between placebo and BMMNC groups.

^eTrends toward differences between placebo and BMMNC groups.

Discussion

This report extends our initial results, showing acute administration of autologous BMMNC after severe traumatic brain injury in children prevents long-term WM loss, preserves CC microstructural indices and reduces ICU metrics of duration/intensity. The data also confirm previous safety findings and logistical feasibility/exportability of the protocol. Additionally, these data replicate previous findings in our phase 1 paediatric clinical trial^6,7 and adult phase 1/2a clinical trial.⁵ In contrast to the earlier studies, our current trial was performed using a prospective, randomized, double-masked design at two sites. As this was a phase 2 trial, it was designed around exploring imaging end points, as opposed to functional outcomes as would be done in a pivotal phase 3 study. As such, the principal planned volumetrics and microstructural data analyses are compelling in that they replicate previous phase 1 data and, moreover, that untreated patients showed a similar atrophy effect as other studies.

Imaging as the primary outcome

We evaluated the impact of BMMNC infusion on brain structural indices since these are, globally, the most representative of biological impact, in contrast to only studying a broad functional outcome, such as GOS-E Peds. Our rationale was that traumatic brain injury lesion locations are often diffuse, with heterogeneity among patients, and location can impact outcome in a volume-independent manner. For example, a small lesion in a critical region of the brain can result in severe disability, while the same volume of lesion in a non-eloquent region has minimal impact. This basic conundrum of lesion location mismatched with lesion size creates functional heterogeneity of patients in clinical trials based on their initial functional severity (GCS). Since therapy impacts only the injured tissue or the response to injury, it is rational to directly measure the tissue outcome instead of functional outcome alone. As BMMNC infusion is thought to impact global neuroinflammatory responses, and we previously showed potential impacts on structural preservation, we chose structural indices as our primary outcomes. Structural preservation has repeatedly been shown to correlate with neurocognitive outcomes^31,33,47-49 and limits the impact of injury heterogeneity. However, while heterogeneity can be overcome with massive trials, this was impractical at this stage,⁵⁴ which is a limitation of our trial.

Global white matter preservation

Based on our previous work with the adult traumatic brain injury population⁵ and supported by preclinical data from our group and others,^55,56 we evaluated several global volumetric measures over time and compared BMMNC versus placebo. Our analyses included five regions of interest based on segmented tissue types from the 3D T1-weighted images acquired: cerebral WM, total GM, supra-tentorial, supra-tentorial (excluding ventricular CSF) and CC. Of these five regions of interest, a significant treatment effect was observed in cerebral WM, with the BMMNC group demonstrating 50.9% less volumetric loss than the placebo group (P = 0.008) at 1-year post-injury.

Progressive WM loss after severe traumatic brain injury occurs over time. The observation of WM but not GM loss in children in our trial is consistent with other reports.⁵⁷ Additionally, the correlation among global GM and WM volumetric loss and functional outcomes has also been demonstrated previously.^33,47 For example, Loane et al.³⁷ demonstrated progressive WM loss over time in a rodent traumatic brain injury model, and this correlated and was co-localized with chronic microglial activation. We have previously demonstrated an impact of BMMNC on microglial activation in a preclinical model,⁴ and our clinical data support modulation of microglial activation as a potential mechanism of action, as manifested by WM preservation. WM volume should increase in the developing brain,⁵⁸ but the volumes are small relative to degenerative losses over the 1-year study period; thus, further analysis of an observed/expected ratio of WM would be of limited utility.⁵⁹ Currently, we are evaluating the degree, location and impact of chronic microglial activation after traumatic brain injury using PET-MRI and radioligands to the translocator protein receptor.

A limitation of our study is the lack of measurements of microglial activity, as this is the putative therapeutic target of this treatment strategy. Our ability to measure this acutely/subacutely after injury is limited due to the extensive imaging time for a critically ill patient and coordinating the logistics of a short half-life isotope with MRI scanning and ICU care. However, other groups have shown loss of specific WM tracts and correlated these with functional outcomes, CC atrophy and/or GM loss.^30-33,47 There have been varying reports on microstructural metrics as well.^60-63 Undoubtedly, these represent the heterogeneity of traumatic brain injury patterns, secondary inflammatory responses and methods of measurement.

Corpus callosum microstructure: midsagittal connectivity metrics of the corpus callosum

The diffusion tensor imaging (DTI) sequence we used was 2.7 mm isotropic and acquired with 30 directions uniformly distributed across the spherical gradient schema (b = 1000 mm²/s). Deterministic tractography methods were used to reconstruct the connectivity of the CC at each time point. Both the number of streamlines and the mean length of streamlines were qualitatively captured via screenshots. Additionally, these features were also documented quantitatively. CC streamlines are highly variable; thus, the 91.4% preservation of CC connectivity at 1 year in the BMMNC group relative to the placebo group did not meet statistical significance (P = 0.055). Had it reached statistical significance, it would also corroborate the significantly preserved total cerebral WM volume in the BMMNC group. Interestingly, the volume of the CC did not significantly differ between groups at any time point. However, previous work has documented a protracted development of CC volumetric loss over time (e.g. several years),⁵⁸ and our final time point was 1-year post-injury. We would like to re-image these patients chronically post-injury to analyse these parameters.

Microstructural metrics: fractional anisotropy and diffusivity (mean, axial and radial)

There was a significant reduction in AD in the CC in the BMMNC group compared to placebo at both the 6-month and 1-year time points. These data are consistent with previous studies, where AD was elevated over time in patients with moderate and severe traumatic brain injury.^60,64 There were strong trends in reduction of MD in the CC in the BMMNC group compared to placebo at both the 6-month and 1-year time points. There were no statistically significant changes in FA. These data suggest disrupted axonal integrity but not a concomitant change in myelination within the body of the CC.

Atrophy occurs as function improves

For the majority of patients, brain volume decline occurred concurrently with clinical improvement. This paradoxical change points to plasticity and/or other mechanisms that restore function despite macroscopic degeneration. Global functional improvement is multifactorial and probably represents improvements in the metabolic homeostasis of CNS, weaning from medications, recovery from non-CNS injuries, environmental enrichment/improved sleep-wake cycles and broad resolution of systemic inflammation. The time course of improvement also coincides with oedema resolution that requires more time than is generally appreciated.

Functional outcome measures

GOS-E Peds was obtained as global outcome measures. Our data show long-term improvement in the dichotomized GOS-E Peds. Using this outcome in a pivotal trial would require 144–156 patients depending upon randomization ration and loss to follow-up. These scores provide an accepted metric, are easily measured and can be benchmarked to other studies. We have been critical of this outcome measure since many factors can influence the scoring. These include access to rehabilitation, caregiver bias and/or comfort with the specific activities that drive the score. Specifically, the scores are obtained from the patient's parents/caregivers and are not a direct measurement per se. In theory, these factors should be equally distributed across treatment groups; however, they create a signal:noise problem, especially in smaller trials. Specifically, patients likely will not move from the bad to good category with any treatment if deaths are included in the analysis. Deaths occur in 50% of the severe traumatic brain injury population in adults. Considering survivors alone, there is typical improvement, as noted in our study.⁶⁵ Also, compounding the lack of difference between groups was a smaller proportion of bad outcomes in our current trial than some previous reports with similar cohorts.⁶⁶

Safety

The data from our study agreed with our phase 1 paediatric trial^6,7 and adult phase 1/2a trial⁵ and confirmed that bone marrow harvest in the acute setting after severe traumatic brain injury did not alter haemodynamics or ICP in any deleterious manner. Further, in our current trial, we found no significant infusional toxicity, using PELOD and Murray scores as the primary measure of organ injury.

Monitoring for venous thromboembolic events related to infusion (particularly pulmonary) of any clinical cellular therapeutic should be incorporated into therapeutic protocols due to the potential for tissue-factor-mediated thromboses.^67,68 We expected relatively lower risk of pulmonary toxicity with BMMNCs, as they are smaller than mesenchymal stromal cells (5–8 μm for BMMNCs versus 15–25 μm for mesenchymal stromal cells, with capillary diameter of about 7 μm for BMMNCs) and have lower tissue factor expression. There was a weak dose-related toxicity trend in the DLT analysis (Supplementary Table 4; low dose 33%; high dose 44%); similar findings were noted in our previously published adult TBI study.⁵

Decrease in therapeutic intensity

The measures of therapeutic intensity we evaluated were ICU days, ventilator days and ICP monitor days; PILOT scores were also assessed as a composite measure of therapeutic intensity. Treatment with autologous BMMNCs was associated with a clinically relevant almost 3-day (23%) reduction in ventilator days, 1-day (16%) reduction in ICP monitoring days and 3-day (14%) reduction in ICU days. As the treating intensivists were masked to randomization, these ICU data are compelling in terms of treatment effect size, and the data support the treatment effect to reduce ICP observed. While clinically relevant, the reductions were not statistically significant due to high variability. ICP measurements as an outcome measure are not useful without continuous waveform capture to quantify the total ICP burden (Intensity × Time); we did not have the capability to capture those data.⁶⁹ Additionally, these data are clinically relevant from a patient perspective, as any reduction in these three parameters leads to increased patient comfort.

While the placebo group had a larger PILOT AUC (more intense treatment) compared to BMMNC, this was not significantly different. The PILOT score is a standardized and validated measure used to compare overall therapeutic intensity in severe paediatric traumatic brain injury. The score is calculated every 24 h following initial trauma⁴³ but did not show a difference in our study. The rationale for using PILOT scores, as opposed to ICP burden alone, to interpret these critical care metrics is related to the common clinical scenario in which two patients may have similar ICPs but different levels of therapeutic interventions to achieve those pressures. Meaning, a patient with an ICP of 13 mm Hg requiring sedation and high-dose hypertonic saline infusion is clinically different than a patient with an ICP of 13 mm Hg requiring no ICP-directed treatment. A limitation of the current PILOT score is a ceiling effect for higher-intensity hypertonic infusions out of the range used during the early 2000s. We often exceeded these peaks, and the PILOT score does not capture that intensity.

Other limitations

Limitations to our study have been discussed above. Further limitations include our small sample size (33 completed 1-year visits). Additionally, there are potential confounders in terms of resuscitation impact on traumatic brain injury outcomes. Our centres use liquid plasma and whole blood for trauma resuscitation. These factors may impact tissue preservation.⁷⁰ Randomization should control for these practices, but in a small study, it is possible to have some influence on outcomes.

Summary

Autologous BMMNC delivered within 48 h after a severe traumatic brain injury in paediatric patients is safe and logistically feasible. DT-MRI showed BMMNC infusion is associated with mitigation of half of the WM loss typically seen after severe traumatic brain injury at 1 year. Treatment was also associated with a statistically and clinically meaningful improvement in acute outcomes, including an almost 3-day reduction in need for mechanical ventilation and a 16% decrease in ICP monitoring days. Microstructural metrics demonstrated a preservation of axonal integrity in the CC and possible preservation of connectivity. While our study here shows tissue preservation potentials of autologous BMMNC, regulatory agencies require additional data showing improvement in how a patient feels or functions, as well as survival rate, in order to approve the treatment for widespread use in traumatic brain injury patients. Thus, the next step would be a phase 3 trial with GOS-E as a primary outcome, with imaging parameters also evaluated as secondary outcomes.

Data availability

The data that support the findings from this study are available from the corresponding author upon reasonable request and in compliance with institutional policy.

Funding

National Institutes of Health grant R01NS077963 (C.S.C.); National Institutes of Health grant T32GM008792 (C.S.C.); Glassell Family Foundation (C.S.C., F.T.); Mission Connect/TIRR Foundation; Lloyd Bentsen Stroke Center Award (C.S.C.). Brown Foundation (C.S.C.); Evelyn Griffin Stem Cell Research Laboratory (F.T.).

Competing interests

C.S.C. and UTHealth have equity/royalty interest in Cellvation, Inc. C.S.C. serves on the Scientific Advisory Board for CBR, Inc. and has sponsored research with CBR, Inc., HopeBio, Inc. and Athersys, Inc. All other authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.