Association of tumor growth rates with molecular biomarker status: a longitudinal study of high-grade glioma

Patient demographics

Table 1 listed the clinical characteristics and the relevant subtypes according to the 2016 WHO classification [1]. A total of 109 patients (66 men and 43 women) were enrolled in the present study. The numbers of patients diagnosed with WHO tumor grade III and IV were 59 and 50, respectively. The median age at the time of tumor detection on the first MRI examination was 48 years (IQR [interquartile range], 35–57 years). All patients received at least twice T₂-weighted images (T₂WI) scans before surgery, 20 received ≥3 times, while 56 patients received at least twice contrast-enhanced T₁-weighted image (CE-T₁WI) scans. The median interval time between the first and the last preoperative MRI examinations was 41 days (IQR, 22.7–114.8 days). The median initial mean tumor diameter (iMTD) was 38.6 cm³ (IQR, 30.0–47.4 mm). Only 44 patients were available to be allocated into the non-not otherwise specified (NOS) subtype because others underwent surgery before 2016 and did not undergo diagnostic molecular testing.

Evaluating tumor growth rate

Multiple growth rates were calculated based on tumor volume and mean tumor diameter (MTD) (Table 2). The equivalent volume-doubling time (eVDT) of the contrast area (based on CE-T₁WI) versus tumor entity (based on T₂WI) was 39.8 days versus 63.4 days, and 61.1 (IQR, 30.8-114.1) versus 40.37 (IQR, 11.7-76) mm/year in median volume-doubling time (VDE). Linear mixed-effects models (LME) with random intercepts was used to evaluate the growth rate of low-grade gliomas (LGGs) and proved that MTD grows linearly [7, 12]. In this study, we considered all the clinical biomarkers as fixed variables and found that iMTD has independent fixed effects (p < 0.01, Supplementary Table 1). Thus, for evaluating eVDEs, iMTD was introduced as a fixed variable into LME. Comparing this with the previous method, it was found that the new method had better prediction accuracy than the previous one (p < 0.01, Supplementary Table 5) [7, 12]. As for eVDE, tumor entity grew at a speed of 51.6 (95% confidence interval [CI], 41.5-61.0) mm/year, and the enhanced region grew by 64.3 (95% CI, 47.8-80.7) mm/year. The eVDEs for each patient were then fitted and aligned with MTD evolution over time (Figure 1).

Association of tumor growth rate with clinical and molecular biomarkers status

To avoid confounding the cohort effects of the clinical biomarkers, we first separately introduced each biomarker as an interaction term and then introduced the significant ones together into interaction terms. Only the WHO grade was an independent factor of the clinical biomarkers, and higher grade (WHO grade IV) was associated with increased tumor growth rate (+27.5 ± 9.8 mm/year, p = 0.005, Supplementary Table 2). having introduced iMTD as a fixed effect and the WHO grade as an interaction term, the molecular biomarkers were then introduced separately as interaction terms into the univariable analysis. As a result, telomerase reverse transcriptase (TERT) promoter mutation C250T (+ 52.4 ± 25.7 mm/year, p = 0.04) was significantly associated with increased tumor growth, while O-6-methylguanine-DNA methyltransferase (MGMT) promoter methylation (-37.4 ± 17.6 mm/year, p = 0.03) was significantly associated with decreased tumor growth. High α-thalassemia X-linked intellectual disability (ATRX) expression (+ 31.6 ± 16.2 mm/year, p = 0.05, score 3/4 versus 0-2); Ki67 expression (+ 20.1 ± 11.3 mm/year, p = 0.08; score 3/4 versus 0-2); and isocitrate dehydrogenase 1 (IDH1) mutation (-28.7 ± 15 mm/year, p = 0.06) were marginally significantly associated with tumor growth (Figure 2, the results of other molecular biomarkers are shown in Supplementary Table 3). These significant and marginally significant biomarkers were introduced together as interaction terms into multivariable analysis, and the results exhibited that only the WHO grade (+19.1 ± 10.5, p = 0.07) showed marginal independence (Table 3).

To further investigate the association of molecular biomarkers with tumor growth, combinations of molecular biomarkers were introduced into interaction terms. After adjusting for WHO grade, ATRX, Ki67, and MGMT, the IDH1, TERT, and 1p/19q molecular groups were independently associated with tumor growth. HGGs with TERT promoter mutation (C250T or C228T) only had a significantly faster growth rate than other groups. After adjustment for WHO grade, ATRX, Ki67 and TERT, the molecular groups of IDH1 and MGMT were independently associated with tumor growth. HGGs with IDH1 wild type and MGMT promoter methylation had a significantly faster growth rate than other groups. After adjustment for WHO grade, MGMT, Ki67, and TERT, molecular group of IDH1 and ATRX were associated with tumor growth in univariable analysis but not in the multivariable model. (Table 4, details of the WHO grade and other molecular biomarkers are shown in Supplementary Table 4).

Comparison of the growth rate in different contrast enhancement types

eVDEs were compared in patients with at least twice CE-T₁WI scans (n = 54) (Figure 3). Firstly, eVDEs of the complete enhanced HGGs and incomplete HGSSs were compared based on CE-T₁WI (tumor enhanced area) and T₂WI (tumor entity), respectively. The results showed no significant difference in eVDEs in contrast to the enhanced type (p > 0.05). Then, eVDEs of the tumor enhanced area and tumor entity were compared in different contrast enhancement types. We found that tumor enhanced area (75.4, 90% CI, 51.8-99 mm/year) showed significantly higher growth rate than tumor entity in incomplete enhanced HGGs (46.7, 90% CI, 28.6-64.8 mm/year, p = 0.006); however, there was no significant difference in complete enhanced HGGs (p > 0.05).

Patient selection

We retrospectively reviewed the clinical information and imaging data from patients with gliomas who underwent primary surgical treatment between January 2008 and March 2019. The inclusion criteria were as follows: 1) age ≥18 years at diagnosis, 2) two or more MRI examinations were performed prior to surgery, 3) no chemotherapy or radiotherapy was administered prior to surgery, and 4) WHO grade III or IV glioma was confirmed histopathologically. To avoid bias, patients for whom sequential MRIs were performed at intervals of less than 14 days were excluded from this study. A total of 109 patients with HGGs were finally included.

Standard protocol approvals, registrations, and patient consents

All clinical information was retrospectively collected from the institutional medical database and the retrospective analysis of this study was approved by the local institutional review board.

Magnetic resonance imaging data acquisition

For most patients, MRI scans were obtained using a Magnetom Trio 3T scanner (Siemens AG, Erlangen, Germany). In other cases, imaging data were acquired using a Magnetom Verio 3T scanner (Siemens AG, Erlangen, Germany). T₂WI were obtained with the following imaging parameters: TR = 5800 ms; TE = 110 ms; field of view = 240188 mm²; flip angle = 150°; and voxel size = 0.6×0.6×5 mm³. Gadopentetate dimeglumine (Ga-DTPA injection, Beijing, Beilu Pharma) was injected intravenously at a dose of 0.1 mM/ kg, and post-contrast T1-weighted images were collected after contrast injection. T1-weighted images were obtained with the following parameters: TE = 15 ms, TR = 450 ms, and slice thickness = 5 mm. The contrast-enhanced area included the contrast area and the necrotic region, marked on the CE-T₁WI. The brain lesions of each patient were manually segmented by two neurosurgeons using the free access software MRIcro (http://www.mccauslandcenter.sc.edu/mricro/), a senior neuroradiologist determined the lesion border if a discrepancy of more than 5% was observed.

IDH1 mutations and MGMT promoter methylation

IDH1 mutations were identified using DNA pyrosequencing [36]. In brief, a QIAamp DNA Mini Kit (Qiagen) was used to isolate genomic DNA from frozen tumor tissue samples. The genomic region spanning the wild-type R132 of IDH1 was analyzed by amplifying a 75-base pair (bp) fragment with the following primers: 5′-GCTTGTGAGTGGATGGGTAAAAC-3′ and 5′-biotin-TTGCCAACATGACTTACTTGATC-3′. Duplicate PCR analyses were performed in 40 μL reaction tubes containing 1 μL each of 10 μM forward and reverse primers, 4 μL of 10 × buffer, 3.2 μL of 2.5 mM dNTPs, 2.5 U HotStar Taq (Takara), and 2 μL of 10 μM DNA. The PCR conditions were as follows: 95°C for 3 minutes, 50 cycles at 95°C for 15 seconds, 56°C for 20 seconds, 72°C for 30 seconds, and then 72°C for 5 minutes (ABI PCR System 9700; Applied Biosystems). Single-stranded DNA was purified from the PCR products and pyrosequenced with a PyroMark Q96 ID System (Qiagen) using a 5′-TGGATGGGTAAAACCT-3′ primer and an EpiTect Bisulfite Kit (Qiagen).

The methylation status of the MGMT promoter was also detected using DNA pyro-sequencing as previously reported [15, 39].

TERT promoter mutation

Mutations of the TERT promoter were identified by polymerase chain reaction (PCR) and Sanger sequencing [15]. Sequences covering genomic mutational hotspots in the TERT core promoter region (nucleotide positions 1,295,228 [C228T] and 1,295,250 [C250T]) were amplified using nested PCR with reference to the human genome reference sequence (grCh37 February 2009; http://genome.ucsc.edu/). PCR was carried out in a total volume of 10 μl containing 1 μl DNA (10–50 ng/μl), Platinum Taq DNA polymerase (1 unit), 1 μl of 10X PCR buffer (Invitrogen, Carlsbad, CA, USA), 1.0 mM MgCl₂, 0.1 mM of each dNTP, 1% (v/v) dimethyl sulfoxide, and 0.25 mM of each primer. Amplified products were purified using the Illustra ExoProStar system (GE Healthcare, Buckinghamshire, UK) to remove any unused primer and were then subjected to direct sequencing with a BigDye Terminator cycle sequencing kit (Applied Biosystems, Foster City, CA, USA) on an ABI 3100 PRISM DNA sequencer. Before sequencing, the quality of all PCR products was checked via electrophoresis on 2% agarose gels.

Detection of 1p/19q codeletion

Representative tumor areas were marked on hematoxylin and eosin-stained sections. The corresponding areas were identified on paraffin blocks, and new 4 μm sections were prepared. The material was deparaffinized with xylene, incubated with 0.3% pepsin in 10 mM HCl at 37°C for 10 minutes, and denatured at 85°C for 10 minutes. Dual-color fluorescence in situ hybridization was performed using LSI probe sets for 1p36/1q25 and 19q13/19p13 (spectrum orange-labeled 1p36 and 19q13 probes; spectrum green-labeled 1q25 and 19p13 probes; and Vysis) and evaluated in at least 200 non-overlapping nuclei with intact morphology.

Immunohistochemical staining

The details of the immunohistochemistry performed was included in the supplementary files. Briefly, formalin-fixed tumor tissues were dehydrated in ethanol and embedded in paraffin. Five-micron-thick sections were prepared, and immunohistochemical staining was performed using antibodies from Zhongshan Gold Bridge Biotechnology of ATRX (1:100 dilution; ZA-0016), primary glial fibrillary acidic protein (GFAP; 1:100 dilution; ZM-0118), oligodendrocyte transcription factor (Olig-2; 1:100 dilution; ZA-0561), topoisomerase II (TOPO2; 1:100 dilution; ZM-0245), P170 (1:100 dilution; ZM-0189), matrix metallopeptidase 9 (MMP9; 1:100 dilution; ZA-0562), glutathione S-transferase π (GST-π; 1:100 dilution; ZM-0110), Ki67(1:100 dilution; ZM-0167), MGMT(1:100 dilution; ZM-0461), epidermal growth factor receptor (EGFR; 1:100 dilution; ZA-0505), vascular endothelial growth factor (VEGF; 1:100 dilution; ZA-0509), phosphatase and tensin homolog (PTEN; 1:100 dilution; ZA-0635) and p53 (1:100 dilution; ZM-0408), according to the protocols.

For the histopathological scoring, the sections were reviewed by two neuropathologists who were blinded to the clinical data. Staining was scored on a 5-point scale ranging from 0 to 4 as follows: 0 = no or rare occurrence of staining, 1 = 10% of cells positively stained, 2 = 10-30% of cells positively stained, 3 = 30-60% of cells positively stained, and 4 = over 60% of cells positively stained. To obtain a sample size of the subgroups that would meet the statistical requirements and determine a meaningful segmentation point of dichotomy for each biomarker, cutoffs were defined. Pathologists who conducted the immunohistochemical analyses were blinded to the clinical and molecular information.

Assessment of inherent tumor growth

The tumor volume (V) was calculated based on the segmented tumor region drawn on transverse T₂WI (Figure 1) using MATLAB (version 2014a, The MathWorks Inc., MA, USA). The contrast-enhanced area was based on CE-T₁WI. The growth rate of HGGs can be assessed using VDT, SGR, eVDT, velocity of diameter expansion (VDE), and equivalent VDE (eVDE). VDT was calculated based on tumor volume [(VDT = ΔT × log 2/(logV₂ – logV₁), where V₁ represents the tumor volume at the first MRI examination, V₂ represents the tumor volume at the most recent MRI examination prior to surgery, and ΔT represents the time interval]. SGR was calculated based on VDT (SGR = ln 2/VDT), and eVDT was calculated using mean SGR (eVDT = ln 2/SGR). SGR is considered to yield a highly symmetrical distribution, while eVDT is considered to yield a more precise estimate of the true growth rate in the population than the median VDT. In addition, VDE was estimated by the linear regression of MTD, (MTD = (2 × V)^1/3) for each patient over time [11].

To characterize changes in MTDs over time and their association with clinical and molecular biomarkers, LMEs were used for the longitudinal data [16, 40]. LME provided more precise predictions of MTD evolution over time and without confounding cohort effects. We used the following formula in this study:

$MT{D}_{ij}={\beta }_0+{\beta }_1\times T_{ij}+{\beta }_m\times I_m+{\alpha }_{1i}+{\alpha }_{2i}\times T_{ij}+{\epsilon }_{ij}$

MTD_ij denotes the MTD for patient i at time of observation j. I = (I₁, I₂, … I_m) are the fixed effects of biomarkers. α and β represent the coefficients of random effects and fixed effects, respectively. β₁ represents eVDE. T_ij represents the time of observation j from first observation for patient i. ε_ij is the residual term. Considering the inter-patient variations, fitted eVDE for patients i can be described as (β₁ + α_2i).

Biomarkers with significant fixed effects were introduced in LME. To describe the change in eVDE associated with the status of I = (I₁, I₂, … I_n), interaction terms were introduced in LME, as presented below:

$\begin{array}{c}MT{D}_{ij}={\beta }_0+{\beta }_1\times T_{ij}+{\beta }_m\times I_m+{\beta }_n\times (I_n\times T_{ij})+{\beta }_{n+1}\times I_n\\ +{\alpha }_{1i}+{\alpha }_{2i}\times T_{ij}+{\epsilon }_{ij}\end{array}$

Thus, the change in eVDE (increased or decreased) associated with the factor I_n are represented as β_n.

Statistical analyses

Statistical analysis was conducted using MATLAB 2014a. To select the fixed variables represented by the significant association with MTD, clinical biomarkers such as age, gender, WHO grade, iMTD, interval time between the first and last MRI examinations, number of MRI examinations, cortisol therapy, contrast enhancement type, tumor-edema interface (clear or blur), number of lobes involved and brain structures involved (frontal lobe, parietal lobe, occipital lobe, temporal lobe, insular lobe, stem, thalamus, cerebellum and ventricle) were introduced, separately into LME. Then the clinical biomarkers with significant fixed effect were introduced together, and fixed variables were selected from significant biomarkers. To select the clinical variables that were significantly associated with eVDE, clinical biomarkers were then introduced into an interaction term alone with time, and then those with significant interaction coefficients were introduced together. Thus, we found the independent clinical biomarkers associated with eVDEs. These variables and time, along with single molecular biomarkers were then introduced into interaction terms in univariable analysis to determine the significance of the molecular biomarkers associated with tumor growth. To find independent factors of eVDEs, we introduced all the significant and marginally significant biomarkers into the interaction terms in multivariable analysis.

For biomarkers with unknown or NOS group, we first evaluated the NOS group with the non-NOS group to determine whether the subgroups were distributed inconsistently between those two groups. Only biomarkers with consistent distribution were taken into consideration. Molecular biomarkers, examined by immunohistochemical staining were allocated to different subgroups according to their expression levels (low versus high, the cutoff was selected by the significant p-values, Supplementary Table 3). A p value < 0.05 was considered to be significant and p value between 0.05 and 0.1 was considered to be marginally significant [41, 42].

Data availability statement

Anonymized data will be shared by request from any qualified investigator.