The role of ADC values within the normal-appearing brain in the prognosis of multiple sclerosis activity during interferon-β therapy in the 3-year follow-up: a preliminary report

Predictors of multiple sclerosis (MS) activity during disease-modifying treatment are being extensively investigated. The aim of this study was to assess the prognosis of NEDA (no evidence of disease activity) status during IFN-β (interferon-β) treatment, using apparent diffusion coefficient (ADC) measurements obtained at initial MRI (magnetic resonance imaging). In 87 MS patients treated with IFN-β, ADC values were calculated for 13 regions of normal-appearing white and grey matter (NAWM, NAGM) based on MRI performed with a 1.5 T magnet before (MS0, n = 45) or after one year of therapy (MS1, n = 42). Associations were evaluated between ADC, conventional MRI findings, demographic and clinical factors and NEDA status within the following 3 years using logistic, Cox and multinomial logistic regression models. NEDA rates in the MS0 group were 64.4%, 46.5% and 33.3% after the 1st, 2nd and 3rd year of treatment, respectively and in MS1 patients 71.4% and 48.7% for the periods 1st–2nd and 1st–3rd years of treatment, respectively. ADC values in the NAWM regions contributed to loss of NEDA and its clinical and radiological components, with a 1–3% increase in the risk of NEDA loss (p = 0.0001–0.0489) in both groups. ADC measurements may have an additional prognostic value with regard to NEDA status.


Patients with clinically defined relapsing-remitting course of MS, according to the 2010 revision of McDon-
ald's criteria 30 ; 2. IFN-β as the only immunomodulatory treatment applied; 3. Availability of brain MRI examinations with appropriate DWI sequence performed before or after 1 year of treatment with IFN-β in the tertiary referral centre; 4. Up to 45 cerebral hyperintense T2/FLAIR lesions visible on conventional MRI; 5. No visible changes on conventional MRI sequences within the white and grey matter within regions of interest for ADC measurements.
All MS patients who did not fulfil the criteria mentioned above were excluded, which is depicted in the flow-chart (Fig. 1).
The patients were divided into two groups according to the timing of the initial MRI: • MS0 group (n = 45)-1st MRI before starting treatment = true baseline MRI; • MS1 group (n = 42)-1st MRI after 1 year of treatment with IFN-β and no available previous MRI scans = ontreatment MRI as baseline.
Demographic, clinical and conventional MRI characteristics of the sample are shown in detail in Table 1. NEDA definition and study outcomes. The main endpoints of our study was a NEDA failure, which was assessed after each year of IFN-β treatment in the MS0 group and within 1st-2nd and 1st-3rd year intervals for the MS1 group. In our study NEDA was defined as having no clinical relapses and/or no disability progression ("clinical" NEDA) and no new/enlarging or no gadolinium enhancing lesion(s) ("MRI" NEDA) 24 . A relapse was denoted as a monophasic (sub)acute clinical episode with symptoms and signs of neurological deficit, with a duration of at least 24 h, with or without recovery, and in the absence of fever or infection 31 . Since progression of the disability was observed only temporarily during relapses and in none of the patients sustained disability beyond relapses was noted, these clinical aspects of NEDA were not rated separately 17,21 . New and enlarged lesion(s) were assessed on T2/FLAIR and T1 post-contrast sequence 32 . The beginning of treatment with IFN-β was a start point for evaluation NEDA in the MS0 group with 1st MRI before therapy. In the MS1 group with 1st MRI after 1 year of treatment none of the patients had a relapse in the 1st year of treatment but history of new lesions formation was unknown in this period, which precluded us from using NEDA from the beginning of Scientific RepoRtS | (2020) 10:12828 | https://doi.org/10.1038/s41598-020-69383-3 www.nature.com/scientificreports/ treatment-consequently in this group the start point for NEDA assessment was set at the 1st MRI. Three-years follow-up means that patients in both groups were clinically observed from the beginning of IFN-β treatment until 3 years had passed. However, because of different time points for starting NEDA assessment, the total length of MRI follow-up was at least 3 years for the MS0 group and 2 years for the MS1 group (from the 1st MRI after 1 year of treatment to finishing the 3-year follow-up procedure). Majority of the patients were assessed clinically and radiologically at least every 12 months from the onset of treatment until finishing the 3-year follow-up, unless a few of them omitted the last MRI control scan. Endpoints for the MS0 group with 1st MRI before treatment are provided first and analogous endpoints for the MS1 group are depicted in round brackets, with subdivisions according to the statistical model used: Model 1: • loss of NEDA within the 1st year of treatment (not applicable for the MS1 group), • loss of NEDA within the 2nd year of treatment (between the 1st and 2nd years of treatment), • loss of NEDA within the 3rd year of treatment (between the 1st and 3rd years of treatment), • loss of "MRI" NEDA within the 1st year of treatment (not applicable for the MS1 group), • loss of "MRI" NEDA within the 2nd year of treatment (between the 1st and 2nd years of treatment), • loss of "MRI" NEDA within the 3rd year of treatment (between the 1st and 3rd years of treatment), • loss of "clinical" NEDA within the 1st year of treatment (not applicable for the MS1 group), • loss of "clinical" NEDA within the 2nd year of treatment (between the 1st and 2nd years of treatment), • loss of "clinical" NEDA within the 3rd year of treatment (between the 1st and 3rd years of treatment), • treatment of IFN-β shorter than 3 years (the same), • alternative treatment within 3 years of follow-up (not applicable for the MS1 group); Model 2: • time to loss of NEDA within 3 years of treatment (from the 1st MRI to 3 years of treatment), • time to loss of "MRI" NEDA within 3 years of treatment (from the 1st MRI to 3 years of treatment), • time to loss of "clinical" NEDA within 3 years of treatment (from the 1st MRI to 3 years of treatment), • time to IFN-β treatment shorter than 3 years (the same), • time to alternative treatment in 3 years of follow-up (not applicable for the MS1 group); Model 3: • reasons for loss of NEDA (the same).
Potential prognostic factors of loss of NEDA and other related endpoints during IFN-β treatment. Demographic and clinical factors evaluated as potential prognostic indicators of all aforementioned endpoints during IFN-β treatment were as follows: sex, age at onset of the disease, age at the beginning of treatment, age at the 1st MRI, disease duration until the 1st MRI and onset of treatment and baseline EDSS. Investigated potential conventional and advanced radiological prognostic factors included:  www.nature.com/scientificreports/ • number of T2/FLAIR lesions at the 1st MRI-a total number of demyelinating plaques assessed on T2-weighted and FLAIR images, • the presence and number of Gd-enhancing lesions on contrast-enhanced T1-weighted images at the 1st MRI, • presence of demyelinating lesions in the cervical spine, evaluated on sagittal T2-weighted and contrastenhanced T1-weighted images at the 1st MRI, • value of ADC within the thirteen selected areas of NAWM and NAGM (Fig. 2).
In each MS patient the onset and disease duration were determined on the basis of medical records. Baseline disability was evaluated using the EDSS score, by an experienced neurologist.
Radiological parameters were obtained from the 1st MR examinations in MS patients before (MS0 group) or after the first year of treatment (MS1 group) and were assessed by an experienced radiologist, unaware of the clinical data, particularly the presence of future relapses or new lesions in future MR scans. image acquisition. MR examinations were performed with a 1.5 T MR scanner (SignaHdx, GE Healthcare) using a 16-channel HNS coil. The conventional MR imaging protocol for MS patients consisted of axial T1-weighted SE, axial T2-weighted FSE, axial FLAIR, sagittal and coronal T2-weighted FRFSE, axial DWI SE/ EPI sequence and post-gadolinium T1-weighted 3D-FSPGR images. DWI was performed using an axial single shot, spin-echo type echo-planar images with the following parameters: b = 0 and b = 1,000 mm 2 /s (TE 89.9 ms, TR 8,000 ms, FOV 26 cm, matrix 128 × 128), thickness: 5 mm, spacing: 0, time of acquisition: 40 s. We used the same method of image acquisition as described in our previously published paper 33 . image analysis and post-processing. Diffusion-weighted imaging data were transferred to a GE Advantage Workstation 4.6. Functool software, provided by the manufacturer, was used in post-processing of images. ADC values were acquired by manually placing regions of interest (ROI), sized approx. 200 mm 2 , using DFOV = 13.0 cm on ADC maps in the following selected normal-appearing white and grey matter regions:  www.nature.com/scientificreports/ T2-weighted images or FLAIR images were utilised to ensure that regions of interest did not include any demyelinating plaques or to exclude partial volume effect from border areas. We decided to choose many regions of interest in an attempt to obtain a more comprehensive insight in a potentially diffused pathology involving NAWM and NAGM. We used the same methods in image analysis and post-processing as we presented in our previously published study 33 .

Statistical analysis.
For binary outcomes (model 1), logistic regression was applied. Receiver operating characteristic (ROC) curves were used to choose the most appropriate cut-off value for a test of a binary classifier. In further analysis, the accuracy of the test was measured by the area under the curve (AUC) 34 . Sensitivity, specificity, predictive positive value (PPV) and negative predictive value (NPV) of the potential prognostic factors of NEDA loss and its components were calculated (model 1). In model 2 (time to reaching endpoint) Cox regression was used. In model 3 (prediction of reasons for NEDA loss) multinomial logistic regression was applied. The statistical outcomes were expressed by a classical odds ratio (OR), and hazard ratio (HR) together with confidence 95% intervals and p-values.
The false discovery rate correction (Benjamini-Hochberg) was added to adjust p-value. Besides, cross-validation was performed using the Somers' Dxy rank discrimination index.
The computation was conducted in the R statistical platform 35 . P values < 0.05 were considered as statistically significant.

Results
Baseline demographic, clinical and MRI characteristics. Out of a total of 236 patients with a relapsing-remitting course of MS, 87 patients with complete clinical data at baseline and in the 3-year follow-up were included ( Fig. 1), with further subdivision into a group of 45 patients with the initial MRI before treatment (MS0 group) and those with the 1st MRI after 1 year of treatment (n = 42, MS1 group). The whole baseline characteristics of both groups are summarised in Table 1. Two patients in the MS0 group and 6 patients in the MS1 group refused contrast administration at the 1st MRI, therefore they were not included in the assessment of presence and number of Gd-enhancing lesions at baseline. The sustaining of NEDA status was analysed until year 3 of observation if sufficient follow-up data was accessible. In the MS0 group, 2 patients were lost-to-follow in the 2nd year of observation, because of switching therapy due to first line treatment failure, and 9 patients were excluded from analysis in the 3rd year of follow-up due to treatment change or discontinuation. In the MS1 group, 3 patients were excluded for NEDA assessment between the 1st and 3rd year of treatment because of a lack of the MRI scan after 3 years of treatment and/or resignation of therapy (Table 2).

NEDA at 1-, 2-and 3-year follow-up.
In the MS0 group with baseline MRI before treatment at the 1st year of follow-up, 29 of 45 patients (64.4%) sustained NEDA and 16 patients (35.6%) lost NEDA. In detail, 11 patients (24.4%) lost NEDA due to MRI activity as a first finding and 9 patients failed due to clinical activity first. At the two-year follow-up, 43 patients were amenable to assessment, 20 (46.5%) maintained NEDA status, whereas 23 patients (53.5%) lost NEDA, in whom 18 patients (41.9%) did not sustain "MRI" NEDA first and 12 patients (27.9%) lost "clinical' NEDA first. At the end of the 3-year follow-up period, 36 out of the initial 45 patients had appropriate data for analysis. Here, only 12 (33.3%) retained NEDA status, 19 patients (52.8%) presented MRI activity first and 10 patients (27.8%) developed relapses first. In the MS1 group with the 1st MRI after 1 year of treatment NEDA was assessed for periods between the 1st and 2nd and between the 1st and 3rd year of IFN-β treatment, because of a lack of baseline MRI obtained before treatment. Between the 1st and 2nd years of follow-up, 30 patients from 42 (71.4%) maintained NEDA status, 9 patients (21.4%) failed in "MRI" NEDA and 5 patients (11.9%) lost "clinical" NEDA at first. In the period between the 1st and 3rd years of treatment, 39 patients had sufficient clinical and radiological data for evaluation-19 patients (48.7%) sustained NEDA status, while 20 patients (51.3%) lost NEDA, including 14 patients (35.9%) with "MRI" NEDA loss, 11 patients (28.2%) with "clinical" NEDA loss as a first finding. In both groups and in any period of observation, a majority of patients lost NEDA due to failure in one component, which was mostly associated with formation of new lesions ( Table 2).
Best individual prognostic markers of loss of NEDA during treatment with IFN-β. We performed an exploratory multivariate analysis of 24 potentially valuable independent predictors of disease activity according to the NEDA concept in a 1-, 2-and 3-year follow-up.

Model 1-prognostic markers of loss of NEDA and its components.
In the MS0 group with baseline MRI before treatment we found only one statistically significant prognostic marker of NEDA loss until the 3rd year of treatment-the ADC value in NAWM of the left cerebellar hemisphere (ROI 2), OR 1.03, p = 0.038. New lesions formation until the 1st year of treatment was predicted by only one factor as well-the ADC value in pons (ROI 3) with OR 1.02, p = 0.026. The most powerful factor in prognosis of loss of "MRI" NEDA was the presence of Gd-enhancing lesion(s) at baseline scan-patients with this feature were almost at a ninefold higher risk of loss of "MRI" NEDA during 2 years of IFN-β treatment.
In the MS1 group with baseline MRI after 1 year of INF-β treatment we found that an increased ADC value in the 1st MRI in the left frontal NAWM (ROI 9), the right and left frontoparietal NAWM (ROI 10 and 11) and in the right temporal NAWM (ROI 12) predicted loss of NEDA between the 1st and 3rd years of treatment with OR 1.02 (p = 0.017-0.047). ADC values obtained from the right and left frontoparietal NAWM (ROI 10 and 11) and from the left temporal NAWM (ROI 13) were significant factors in prognosis of "clinical" NEDA loss between None of the factors predicted treatment with IFN-β shorter than 3 years or treatment change in both groups. (Table 3, Suppl. Fig. S1). Based on the ROC curves (Suppl. Fig. S2) analysis, accuracy for different cut-off values of the significant indicators of disease activity was evaluated, which is shown in detail in Table 4. The highest accuracy was achieved in prognosis of loss of "clinical" NEDA between the 1st and 2nd year of treatment in the MS1 group for ADC measurements in the right and left frontoparietal NAWM (ROI 10 and 11) and the left temporal NAWM (ROI 13) (AUC 80.3-81.6%, sensitivity 80-100%, specificity 64.9-75.5%) as well as in prognosis of loss of "clinical" NEDA between the 1st and 3rd year of treatment in the same group for the ADC value in the right frontoparietal NAWM (ROI 10) (AUC 81.4%, sensitivity 72.2%, specificity 83.9%). Other aforementioned significant prognostic factors of disease activity in both groups reached moderate accuracy as single measurements (68.4-79.8%). None of the combinations between significant factors gained accuracy over the single measure.
Model 2-prognosis of time to NEDA loss and its components. MS0 patients with baseline presence of the Gdenhancing lesion had a 2.9-fold risk of shorter time to reach loss of NEDA (p = 0.0097) and a 4.8-fold risk to reach loss of "MRI" NEDA (p = 0.0004) in the 3-year follow-up. Time to loss of "MRI" NEDA was also connected with the baseline number of the Gd-enhancing lesion with a 1.25 hazard ratio (p = 0.0048) and an ADC value in pontine NAWM with a 1.01 hazard ratio (p = 0.038). Baseline EDSS increased the risk of reaching loss of "clinical" NEDA up to 1.32 times (p = 0.032). All mentioned factors were associated with a slight to low crossvalidation index (Dxy = 0.17-0.31).
In the MS1 group, measurements of ADC in a few particular locations at the 1st MR after 1 year of treatment played a role in prediction of the time to reach loss of NEDA and together with the number of T2/FLAIR lesions at the 1st MRI in prediction of the time to loss of "clinical" NEDA. All of the ADC values obtained from both frontal NAWM regions (ROI 8 and 9), both frontoparietal NAWM regions (ROI 10 and 11) and from the right temporal NAWM region (ROI 12) were associated with 1.01 HR (p = 0.0056-0.045). The number of T2/FLAIR gave about a 6% higher risk of reaching loss of "clinical" NEDA. These factors were found to have low to moderate

MS1 group-the 1st MRI after 1 year of treatment with IFN-β only-on-treatment MRI as baseline
Loss of NEDA between the 1st and 2nd years of treatment  Table 3. Logistic regression analysis of the association between significant MRI prognostic markers and endpoints related to disease activity (model 1). Presence of the Gd-enhancing lesion at baseline MRI was the strongest predictor of "MRI" NEDA loss in the MS0 group. ADC measurements obtained from the 1st MRI after 1 year of treatment (MS1 group) were most the common significant markers in prognosis of NEDA loss and "clinical" NEDA loss in the subsequent 2 years of IFN-β treatment, with the best performance in a cross-validation procedure, reaching the Somers' Dxy rank discrimination index > 0. 6   Benjamini-Hochberg correction. In analysis of the false discovery rate correction, all provided potential predictors remained statistically significant with the adjusted p value < 0.05-data are summarised in Tables 3  and 5.  www.nature.com/scientificreports/ cross-validation procedure. In the course of cross-validation procedure we found that the best performance with the Somers' Dxy rank discrimination index was achieved in model 1 in MS1 group in prediction of the "clinical" NEDA loss in both evaluated periods (1st-2nd and 1st-3rd year of IFN-β treatment), reaching moderately high index (Dxy = 0.54-0.63) ( Table 3). Apart from that majority of previously described statistically significant prognostic factors were characterised by low to moderate index (Dxy = 0.31-0.49). Details are provided in the last column of Table 3 and 5.

Discussion
The prompt identification of MS patients with insufficient response to disease-modifying treatment is pivotal for selecting candidates to change the therapy, aiming at an individualised treatment strategy for optimal control of the disease 2,4 . The fundamental challenge in the evaluation of potential predictors is the lack of a shared definition of treatment response in MS 2 . Non-response to treatment is usually assessed on the basis of one of three outcomes or their combination: disability progression 5,11,14,16 , occurrence of relapses 11,12 , and/or presence of new/active MRI lesions 19 22 . Nowadays, some modifications to the primary definitions have been made, with addition of the fourth and fifth domain-pathological volume loss of the brain and raised cerebrospinal fluid neurofilament level, respectively 24 . NEDA is increasingly used in real-world datasets 23 , nevertheless there are some flaws in using NEDA as a goal of treatment-the main drawback is that each NEDA component is equal. As a consequence, a new clinically silent lesion in MRI is as important as the occurrence of a severely disabling relapse 36 . In our study we applied NEDA criteria retrospectively in the RRMS cohort treated with IFN-β as a first-line therapy. We demonstrated NEDA rates of 64.4%, 46.5% and 33.3% after the 1st, 2nd and 3rdyear of treatment, respectively in patients with the initial MRI before therapy, as well as 71.4% and 48.7% for periods 1st-2nd and 1st-3rd years of treatment, respectively, in patients with the 1st MRI after 1 year of treatment (Table 2). Higher rates of NEDA maintenance in the second group (MS1) may result from underestimation of the first year of therapy, when it was not amenable to assess the formation of new lesions, due to a lack of MRI data. Besides, we received higher rates of NEDA maintenance in both groups compared to the study of Huhn et al., where rates of 45%, 29%, and 21% were achieved at each analogical year of treatment 23 compared with our MS0 group. Apart from a known individual variability in the course of MS, these differences may result from distinctive inclusion criteria-we analysed patients treated with IFN-β only, whereas in the cited study patients treated with various drugs were included 23 .
Commonly reported predictors or prognostic factors of treatment response include age 5 , sex, disease duration 6 , baseline disability 6,8 , number of relapses during the first year of treatment 9 , presence and/or number and/or volume of new lesions on T2-weighted images or gadolinium-enhancing lesions [8][9][10][11][12][13][14] . In our cohort we found a few relations between some of these factors and risk of NEDA loss. In the MS0 group the presence of Gd-enhancing lesion(s) at baseline scan was the best predictor of "MRI" NEDA loss until the 2nd year of therapy (OR 8.94), time to NEDA loss (HR 2.9) and time to "MRI" NEDA loss (HR 4.77). Apart from that, baseline EDSS predicted time to loss of "clinical" NEDA with a 1.37 hazard ratio. However, all of them reached slight to low validation index (Dxy 0.17-0.36). In the MS1 group, the number of T2/FLAIR lesions at the 1st MRI favoured loss of "clinical" NEDA between the 1st and 3rd years of therapy (OR 1.11, Dxy 0.55) and increased the risk of time to loss of "clinical" NEDA (HR 1.06, Dxy 0.38) (Tables 3 and 5).
As there is a scarcity of data regarding the use of DWI measurements in the prognosis of IFN-β treatment response 16 , we decided to evaluate the usefulness of easily accessible ADC measurements in this role. Involvement of so-called normal-appearing white and grey matter by a pathological process even in early diagnosed MS patients has been commonly investigated 16,27,37,38 and a connection with that detected by advanced MRI techniques facilitated diffusion was made 28 . This observation is mainly explained by three phenomena in the literature: firstly, conventional MRI sequences used at 1.5 T or 3 T may be deficient in demonstrating all demyelinate lesions in both white and grey matter-the use of higher field strength devices enables the detection of more plaques in regions supposed to be normal on conventional MRI 37 . Secondly, diffusion derangements can predate the formation of new demyelinate plaques 38 . Thirdly, even though MS is regarded as a primarily demyelinating disease, the axonopathy occurs parallel beyond the plaques even at the early stage of the disease 38,39 , which may be related to Wallerian degeneration, in which axons traversing focal lesions are damaged even at a far distance from the demyelinating plaque 40,41 . Retrogenesis is the other hypothesis, assuming that late-myelinating fibres are more prone to age-related and disease-related degeneration and consequently, they are more likely to show diffusion alterations first 42 .
According to our analysis, ADC measurements in some regions are connected with a statistically significant higher risk of loss of NEDA and its components. In a group of patients with baseline MRI before treatment, we found that ADC measured in infratentorial white matter regions (ROIs 1-3, cerebellum and pons) predicted loss of NEDA, loss of "MRI" NEDA (model 1) and time to "MRI" NEDA loss (model 2), but with slight to moderate validation coefficient (Dxy 0.18-0.47). Conversely, in patients with the 1st MRI after 1 year of treatment, supratentorial ADC measurements (ROIs 8-13-frontal, frontoparietal and temporal NAWM) escalated the risk of NEDA loss between the 1st and 3rd year of treatment and "clinical" NEDA loss between the 1st and 2nd, reaching the highest accuracy ≥ 80% and the highest performance in cross-validation (Dxy 0.61-0.63) among investigated factors. ADC measurements in these locations predicted also the "clinical" NEDA failure between the 1st and 3rd year of treatment with a little lower accuracy of 77.0-81.4% and validation coefficient (Dxy 0.54-0.63). In model 2, time to loss of NEDA and its "clinical" component was predicted by ADC obtained from the supratentorial NAWM regions as well. Each ADC value in both groups was associated with an increase in appropriate risk (OR or HR) between 1-3% (Tables 3 and 5), which was far lower than the increase of some clinical and conventional MRI factors (e.g. the presence of a Gd-enhancing lesion), but, important to note, an increase in just one step of ADC value (e.g. from 779 to 780 × 10 -6 mm 2 /s) implicated escalation of a risk in 1-3%. Apart from that most other investigated factors were not useable in predicting any of the seventeen proposed outcomes, whereas ADC measurements did.
There are some limitations of our study. Firstly, the study sample is small-we could include only 87 from all 236 patients managed at our institution because of the tight inclusion criteria. Secondly, we had only 45 patients with true baseline MRI before treatment, however we found that ADC measurements in on-treatment MRI are even more commonly and in better degree associated with statistically significant prediction of subsequent loss of NEDA, particularly the loss of "clinical" NEDA. Thirdly, ADC measurements cannot be applied to all MS patients-patients with numerous plaques and plaques in regions of interest are not amenable to this method. On the other hand, we provided some novelty by using ADC measurements in the prognosis of disease activity during IFN-β treatment as a simple and easy tool.

conclusions
In conclusion, our results seem to suggest that the ADC measurements of NAWM, particularly in the frontal, frontoparietal and temporal regions obtained after 1 year of treatment, may contribute to the prognosis of MS activity according to the NEDA concept during interferon-β therapy. This is a relatively easy imaging measure, that might be incorporated into standard follow-up procedures in MS patients. However, further research of this method should be performed before its widespread implementation.