Thalamic alterations remote to infarct appear as focal iron accumulation and impact clinical outcome

Abstract

See Duering and Schmidt (doi:10.1093/awx135) for a scientific commentary on this article.

Thalamic alterations have been observed in infarcts initially sparing the thalamus but interrupting thalamo-cortical or cortico-thalamic projections. We aimed at extending this knowledge by demonstrating with in vivo imaging sensitive to iron accumulation, one marker of neurodegeneration, that (i) secondary thalamic alterations are focally located in specific thalamic nuclei depending on the initial infarct location; and (ii) such secondary alterations can contribute independently to the long-term outcome. To tackle this issue, 172 patients with an infarct initially sparing the thalamus were prospectively evaluated clinically and with magnetic resonance imaging to quantify iron through R2* map at 24–72 h and at 1-year follow-up. An asymmetry index was used to compare R2* within the thalamus ipsilateral versus contralateral to infarct and we focused on the 95th percentile of R2* as a metric of high iron content. Spatial distribution within the thalamus was analysed on an average R2* map from the entire cohort. The asymmetry index of the 95th percentile within individual nuclei (medio-dorsal, pulvinar, lateral group) were compared according to the initial infarct location in simple and multiple regression analyses and using voxel-based lesion-symptom mapping. Associations between the asymmetry index of the 95th percentile and functional, cognitive and emotional outcome were calculated in multiple regression models. We showed that R2* was not modified at 24–72 h but showed heterogeneous increase at 1 year mainly within the medio-dorsal and pulvinar nuclei. The asymmetry index of the 95th percentile within the medio-dorsal nucleus was significantly associated with infarcts involving anterior areas (frontal P = 0.05, temporal P = 0.02, lenticular P = 0.01) while the asymmetry index of the 95th percentile within the pulvinar nucleus was significantly associated with infarcts involving posterior areas (parietal P = 0.046, temporal P < 0.001) independently of age, gender and infarct volume, which was confirmed by voxel-based lesion-symptom mapping. The asymmetry index of the 95th percentile within the entire thalamus at 1 year was independently associated with poor functional outcome (P = 0.04), poor cognitive outcome (P = 0.03), post-stroke anxiety (P = 0.04) and post-stroke depression (P = 0.02). We have therefore identified that iron accumulates within the thalamus ipsilateral to infarct after a delay with a focal distribution that is strongly linked to the initial infarct location (in relation with the pattern of connectivity between thalamic nuclei and cortical areas or deep nuclei), which independently contributes to functional, cognitive and emotional outcome.

stroke, thalamus, iron, neurodegeneration, clinical outcome

Introduction

Post-stroke outcome is complex and includes not only functional outcome (the physical consequences of ischaemic stroke), but also cognitive impairment and emotional disorders that are major causes of disability (Paolucci, 2008; Mijajlovic et al., 2017). Approximately one-third of post-stroke patients will develop cognitive impairment or depression, but the underlying mechanisms remain unclear (Mijajlovic et al., 2017). The main features of the acute infarct lesion, such as its volume, are predictors of functional outcome (Vogt et al., 2012) but are particularly modest predictors of the cognitive (Munsch et al., 2016) and emotional outcome (Carson et al., 2000). This indicates that, beyond the infarct lesion itself, cognitive and emotional impairments may reflect more global brain dysfunction that could result from disruption of central networks and disconnection of remote brain areas (Siegel et al., 2016). Especially, with thalamus being a central hub of several networks, secondary thalamic alterations might be an important contributor to the long-term outcome. Indeed, thalamic alterations remote to infarct initially sparing the thalamus have been described for several years and attributed to secondary degeneration of thalamo-cortical or cortico-thalamic projections (Matthews, 1973; Iizuka et al., 1990; Tamura et al., 1991; Nakane et al., 1997).

Around this concept, one has to consider that the thalamus is not a single structure but is composed of several nuclei interlocked in a complex anatomy (Morel et al., 1997). Each thalamic nucleus connects specific cortical areas and deep nuclei in such a way that it is possible to parcellate the thalamus based on its cortical connectivity (Behrens et al., 2003; Johansen-Berg et al., 2005). Therefore, individual thalamic nuclei should be differentially affected by secondary degeneration depending on the initial location of the infarct. Highlighting such regionality of secondary modifications within the thalamus after stroke might be relevant for recovery because individual thalamic nuclei drive specific functions especially for cognitive and emotional processes (Sherman, 2005). Nevertheless, regional quantification of secondary thalamic degeneration and demonstration of the relation between initial infarct location and affected thalamic nuclei have not been described so far in humans. Similarly, the clinical impact of such degeneration remote to infarct is still unclear.

Experimental studies in animals support the concept of regional heterogeneity of secondary lesions within the thalamus thanks to immunohistological methods or high resolution in vivo and ex vivo imaging that could show thalamic modifications focally located (Justicia et al., 2008; Walberer et al., 2014) and that can vary with the cortical site of the infarct (Langen et al., 2007). In humans, a seminal paper of 30 middle cerebral artery stroke patients showed that T₂ hyperintensity can be focally distributed within the ipsilateral thalamus (Ogawa et al., 1997) 1–12 months after stroke. Nevertheless, such alterations on T₂-weighted images are not specific, might lack sensitivity and are non-quantitative. Furthermore, such qualitative abnormalities were not precisely localized within specific thalamic nuclei, were not linked to the pattern of the initial infarct and their clinical relevance was not addressed (Ogawa et al., 1997). More recently, quantitative MRI methods such as diffusion (Herve et al., 2005), spectroscopy (Li et al., 2011) and PET imaging (Pappata et al., 2000; Gerhard et al., 2005) have been able to quantify secondary modifications, but only within the entire thalamus without providing any regional information.

In preclinical studies (Justicia et al., 2008; Walberer et al., 2014), among histological modifications, iron accumulates remotely within subregions of the thalamus with a dynamic evolution in close relationship with neuroinflammation and neurodegeneration. At the subacute phase, iron staining is localized intracellularly in activated microglia, a surrogate marker of post-stroke neuroinflammation, whereas at the chronic phase, iron deposits are localized extracellularly around plaque-like amyloid deposits and are clearly associated with marked neuronal loss, indicating neurodegeneration. MRI is particularly sensitive to iron accumulation that causes pronounced susceptibility effects and increases R2* relaxation rates (Brass et al., 2006). The transverse relaxation rate R2* is linearly correlated with chemically-determined brain iron concentration (Langkammer et al., 2010). R2* mapping might therefore be capable of quantifying focal thalamic modifications secondary to infarction of connected cortical areas or deep nuclei, at high spatial resolution and with high sensitivity to test the clinical impact of such remote degeneration.

Therefore, the aim of this study was to use R2* as a marker of iron accumulation associated with neurodegeneration to demonstrate that (i) secondary thalamic alterations are focally located within specific thalamic nuclei depending on the initial infarct location; and (ii) such secondary degeneration can contribute independently to the long-term post-stroke outcome.

Materials and methods

Patients

Patients admitted for an ischaemic stroke between June 2012 and February 2015 were enrolled in a prospective cohort study. The study design included initial (24–72 h) and final (1 year) clinical and MRI evaluations. The institutional review board approved the study and written informed consent was obtained from all patients.

Primary inclusion criteria were as follows: (i) patients older than 18 years old; (ii) with a suspected clinical diagnosis of minor to severe supratentorial cerebral infarct [National Institutes of Health Stroke Scale (NIHSS) between 1 and 25]; and (iii) confirmed on diffusion-weighted imaging (DWI) at 24–72 h.

Exclusion criteria were: (i) history of symptomatic cerebral infarct with functional deficit (pre-stroke modified Rankin Scale score ≥1; to measure the impact of the infarct itself and its secondary consequences on long-term outcome without being biased by pre-existing deficits); (ii) infratentorial infarct; (iii) thalamic infarct; (iv) thalamic microbleed or haemorrhage; (v) history of severe cognitive impairment (dementia) or DSM-IV axis 1 psychiatric disorders (except major depression); (vi) coma; (vii) pregnant or breastfeeding females; and (viii) contraindications to MRI.

The study group consisted of 172 patients (121 males, 51 females, Supplementary Table 1) who met the inclusion criteria and for whom a complete set of data at both baseline and 1-year follow-up were available. The flow chart of the study is presented in Supplementary Fig. 1.

Clinical assessment

The NIHSS was recorded at 24–72 h after stroke onset and at 1-year follow-up. At 1-year follow-up, the modified Rankin Scale was recorded to assess the global disability, the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005) and the Isaacs Set Test (Isaacs and Kennie, 1973) were recorded to assess post-stroke cognitive disability, and the Hospital Anxiety and Depression Scale (HADS; Zigmond and Snaith, 1983) to assess post-stroke emotional disability. The level of education was evaluated according to a 5-point ordinal scale adapted from the International Standard Classification of Education (childhood, primary, lower secondary, upper secondary or tertiary education). The pre-stroke cognitive state was also estimated by IQCODE (Informant Questionnaire on Cognitive Decline in the Elderly), which was completed by the patient’s relative at the time of admission to ensure the absence of significant pre-existing impairment that could contribute to the long-term outcome independently from the infarct itself and its thalamic consequences (mean IQCODE score was 3.1 ± 0.3).

MRI

All magnetic resonance examinations were performed on a 3 T scanner (Discovery MR 750w, GE Medical Systems) with a 32-channel phased array head-coil with exactly the same protocol at 24–72 h (initial) and 1-year (final) after stroke. The imaging protocol included a DWI sequence (38 slices; repetition time, 9000 ms; echo time, 76 ms; slice thickness, 4 mm; matrix, 128 × 128; field of view, 240 mm × 240 mm; b values, 0 and 1000 s/mm²). The protocol also included a 3D T₁ inversion-recovery-prepared fast spoiled gradient echo sequence (196 sagittal slices; repetition time, 8.60 ms; echo time, 3.27 ms; inversion time, 450 ms; flip angle, 12°; slice thickness, 1 mm; matrix, 256 × 256; field of view, 240 mm × 240 mm) a 3D fluid-attenuated inversion-recovery (FLAIR) sequence (224 sagittal slices; repetition time, 9000 ms; echo time, 142 ms; inversion time, 50 ms; slice thickness, 1.8 mm; matrix, 288 × 224; field of view, 240 mm × 240 mm) and a 2D multi-echo fast gradient-echo sequence (28 slices; repetition time, 775 ms; eight echo times at 4.3, 8.7, 13.0, 17.4, 21.7, 26.1, 30.5 and 34.8 ms; flip angle, 20°; slice thickness, 4 mm; matrix, 320 × 320; field of view, 240 mm × 240 mm).

Image processing

DWI and FLAIR analyses

Laterality and anatomic location of infarct on initial DWI were recorded by a neuroradiologist blinded from the R2* maps. In patients with bilateral infarcts, the hemisphere with the largest infarct volume on DWI was considered as the infarct hemisphere. Infarct location was quoted as involving or not the cortex (frontal, parietal, temporal, occipital, insular) and/or the deep nuclei (caudate, lenticular). Infarct volumes were quantified by segmentation of the DWI maps using an automatic pre-detection tool based on intensity variation and edge detection available in 3D Slicer (http://www.slicer.org). White matter changes were quoted according to the Fazekas scale on initial FLAIR (Fazekas et al., 1987).

R2* mapping

R2* mapping was performed as previously described (Peran et al., 2007) with ImageJ software (https://imagej.nih.gov/ij) at 24–72 h and 1 year. A mono-exponential signal decay with echo time was obtained by a voxel-by-voxel nonlinear least-squares fitting of the multi-echo T₂*-weighted data [S(t) = S₀.e^−t.R2*; where t = echo time, S = measured signal, R2* = transverse relaxation rate]. R2* maps and the raw T₂* images were inspected by a neuroradiologist blinded from DWI and clinical data to look for focal area of high R2* values within the thalamus and to quote the presence of extra-thalamic haemorrhage (haemorrhagic transformation or intracerebral haemorrhage).

Co-registration

All the images were normalized to the standard MNI152 space atlas using SPM12 software package (Statistical Parametric Mapping, Welcome Trust Center for Neuroimaging, London, UK). For initial DWI and the associated infarct masks, rigid registration of the b0 DWI volume to the native 3D T₁ volume was performed, followed by non-linear deformation of the 3D T₁ to MNI152 space. After an initial rigid registration of the first echo of the multi-echo T₂*-weighted sequence to the native 3D T₁, R2* was normalized to MNI152 space by applying the previous non-linear registration to MNI152 space. The quality of registration was checked by visual inspection.

R2* measurements at initial and final MRI

Because all analyses were based on measurement of R2* as a marker of iron accumulation, we first validated the reliability and reproducibility of this imaging measurement, by preparing a phantom composed of vials filled with solutions of increasing iron concentration [iron (III) chloride (Sigma-Aldrich)] from 0 to 550 mg/l. The phantom was scanned twice (3-day interval) with the 2D multi-echo fast gradient-echo sequence used for the patients.

To be able to measure R2* in vivo accurately within the whole thalamus but also within specific nuclei, we created a thalamic atlas from independent in vivo data collected at ultra-high field strength (Su et al., 2015). Indeed, we previously identified a specific contrast at 7 T that enables a unique delineation of the thalamus and many of the thalamic nuclei (Tourdias et al., 2014). From this approach we generated a thalamic atlas that is therefore the result of expert manual delineations of the thalamus and 12 individual nuclei from a distinct cohort of 29 subjects (even number of healthy controls and patients) scanned at 7 T. This detailed in vivo thalamic atlas was then registered to MNI152 space in such a way that thalamic nuclei, which cannot be accurately identified otherwise at 3 T, can be inferred from the 3 T images of our stroke patients registered to the MNI152 space.

To quantify iron accumulation, we considered the non-infarct side of the brain as an intrinsic control for each patient. We first computed parameters of the R2* histogram [histogram parameters (HP): 5th percentile, median, 95th percentile and full-width at half-maximum] within the whole thalamus ipsilateral and contralateral to infarct. An asymmetry index of R2* (van Etten et al., 2015) for each parameter of the histogram was also calculated with:

$$\text{Asymmetry}{\text{index}}_{\text{HP}}=\text{1}00\times \left(\text{R2}{{\hspace{0.17em}}^{*}}_{\text{HP}}\left[\text{ipsilateral}\right]-\text{R2}{{\hspace{0.17em}}^{*}}_{\text{HP}}\left[\text{contralateral}\right]\right)/\left(\text{R2}{{\hspace{0.17em}}^{*}}_{\text{HP}}\left[\text{ipsilateral}\right]+\text{R2}{{\hspace{0.17em}}^{*}}_{\text{HP}}\left[\text{contralateral}\right]\right)$$

(1)

We will refer to AI₉₅ for the asymmetry index of the 95th percentile of R2* that is the metric representing the highest values and therefore high iron content.

To identify the spatial distribution of high R2* values and its relation with the initial infarct location, we flipped individual final R2* maps so that all infarct lesions were on the right hemisphere and we built an average colour-coded R2* map for the entire population. We assumed that thalamic connections with cortical areas or deep nuclei are symmetric (Behrens et al., 2003) and we considered the non-infarct hemispheres as healthy controls. From final R2* maps, R2* values of ipsilateral and contralateral thalamus were compared using voxel-based analysis and paired sample t-test based on permutation inference with threshold-free cluster enhancement approach (Smith and Nichols, 2009). The resulting t-map was controlled for multiple comparisons using familywise error rate correction to ensure a false positive rate of P < 0.05.

We also considered quantifying the AI₉₅ within individual nucleus. The thalamus can be divided into four nuclear groups (Morel et al., 1997). The main analyses have been conducted by measuring AI₉₅ within the medio-dorsal nucleus (the main nucleus from the medial group) and the pulvinar nucleus (the main nucleus from the posterior group). We primarily focused on these two nuclei because they show the most robust pattern of cortical connectivity among subjects in the literature (Johansen-Berg et al., 2005), because they can be robustly outlined for the atlas (Tourdias et al., 2014) and based on the observation of the final average R2* map described above (see ‘Results’ section). Additional analyses were also conducted with quantification of AI₉₅ within the lateral group (ventral lateral anterior + ventral lateral posterior + ventral posterior lateral) but we were not able to evaluate secondary alterations within nuclei from the anterior group given their small size and the low prevalence of infarcts expected to be connected with them (Supplementary Fig. 2).

Maps of infarct locations associated with high AI₉₅ within each nucleus

To demonstrate the relationship between high AI₉₅ within specific thalamic nuclei (our marker of focal iron accumulation) and infarct location at the voxel level, we used the voxel-based lesion symptom mapping (VLSM) method (Bates et al., 2003). First, a prevalence map of the infarct lesions was built using the segmented DWI maps registered into the MNI152 space and flipped so that all the infarct lesions projected onto the right hemisphere (Supplementary Fig. 2). Then, we used the VLSM method implemented in the non-parametric mapping toolbox included in the MRIcron software package (MRIcron, Version 15.6.2015) (Rorden et al., 2007). For each voxel, patients were divided into two groups according to whether they did or did not have a lesion affecting that voxel. Instead of behavioural scores, we used the final AI₉₅ within the medio-dorsal, pulvinar and lateral group nuclei as our outcome variables. For each voxel, final AI₉₅ within the medio-dorsal, pulvinar and lateral group nuclei, respectively, were thus compared between the lesioned and the non-lesioned group using a Brunner–Munzel rank order test. Only voxels affected in ≥10 subjects have been tested to avoid potential artificial inflation of Brunner-Munzel scores. The resulting Z-score maps were controlled for multiple comparisons using the false discovery rate correction to ensure a false positive rate of P < 0.05 and P < 0.01. In this way, VLSM could map the infarct voxels significantly associated with high AI₉₅ in specific thalamic nuclei. This method allowed us to directly test our hypothesis of focal thalamic alterations determined by the initial infarct location through thalamo-cortical or cortico-thalamic disconnection and, especially, that the medio-dorsal nucleus is perturbed by infarcts involving anterior cortical areas (prefrontal cortex, anterior temporal cortex), that the pulvinar nucleus is perturbed by infarcts involving posterior cortical areas (parietal cortex and posterior temporal cortex) and that the lateral group is perturbed by infarcts involving precentral and postcentral cortex.

Statistical analysis

Statistical analyses were performed with SPSS 19.0 software (SPSS Inc., Chicago, IL, USA). The distribution of all variables was tested for normality with the Shapiro-Wilk test. To demonstrate iron accumulation within the thalamus ipsilateral to infarct, we compared the parameters from the histogram of R2* ipsilateral and contralateral to infarct with a paired sample t-test. Asymmetry indices were also tested by comparing the mean values with 0 (the hypothesis being a symmetric distribution) with a one-sample t-test.

To demonstrate the relationship between focal iron accumulation in specific nuclei and initial infarct location, we compared the final AI₉₅ within the medio-dorsal, pulvinar and lateral group nuclei according to the initial infarct location (involvement or not of the different cortical locations and deep nuclei). In another analysis, the final AI₉₅ within the medio-dorsal, pulvinar and lateral group was dichotomized by defining iron accumulation for values superior to the third quartile. Odds ratios were calculated using simple and multiple logistic regression with backward stepwise analyses, to predict dichotomized AI₉₅ considering infarct location as an explicative variable. Variables expected to influence iron accumulation were included in the simple regression analysis: age (Aquino et al., 2009), gender (Bartzokis et al., 2007), infarct volume, presence of extra-thalamic haemorrhage (Wu et al., 2010). Variables that reached a P-value of < 0.2 were included in the multiple regression analysis.

Finally, we tested the clinical impact of dichotomized AI₉₅ on functional, cognitive and emotional outcomes. For functional outcome, a logistic regression analysis was performed with modified Rankin Scale at 1-year follow-up as the dependent variable and the final AI₉₅ in the entire thalami as explicative variable. We used a sliding dichotomy approach as recommended by Murray et al. (2005). For cognitive and emotional outcomes, logistic regression analyses with fixed dichotomy were performed with the aim to predict poor cognitive [defined as MoCA score ≤ 25 (Lees et al., 2014) or Isaacs Set Test score ≤ 28 (Mehrabian et al., 2015)] or poor emotional outcome [post-stroke anxiety defined as HADS-A score ≥ 8 or post-stroke depression defined as HADS-D score ≥ 8 (Zigmond and Snaith, 1983; Prisnie et al., 2016)]. We chose these thresholds because they had been previously validated as clinically relevant for the diagnosis of cognitive impairment or emotional disorders in stroke patients. To assess the clinical impact of high AI₉₅ within a particular thalamic nucleus on specific cognitive or emotional domains supposed to depend of the integrity of the given nucleus, we repeated the analyses by introducing final AI₉₅ measured within the medio-dorsal and pulvinar nuclei to predict specific alterations of the visuospatial/executive and recall MoCA subscores (≤3/5) for cognition or HADS-A and HADS-D scores (≥8) for emotion. Given the lateralization of many cognitive processes (Munsch et al., 2016), analyses were also tested separately in patients with left hemispheric or right hemispheric infarcts. Associations with a P-value < 0.2 in simple regression were tested in multiple regression with backward stepwise analysis adjusted for age, gender, education level, white matter changes, infarct volume, lateralization and/or location.

Statistical analyses were performed with a type I error set at α = 0.05.

Results

Phantom study: reliability and test-retest variability of R2* measurements

R2* values within the vials showed a strong linear correlation with iron concentration (R² = 0.96; P < 0.0001). Furthermore, the Bland-Altman plot showed that the variability from one session to another was small and measured on average at −0.28 s⁻¹ for the range of R2* observed in vivo in patients (Supplementary Fig. 3).

Thalamic R2* shows delayed regional modifications ipsilateral to infarct

At initial MRI, R2* measurements within the entire thalamus showed no difference between thalamus ipsilateral and contralateral to infarct (Table 1).

Table 1

R2* measurements within the thalami at initial and final MRI (n = 172 patients)

R2* values	Thalamus ipsilateral to infarct		Thalamus contralateral to infarct		P-value	Asymmetry index		P-value
	Mean	95% CI	Mean	95% CI		Mean	95% CI
Initial MRI
Fifth percentile	14.3	14.0, 14.6	14.1	13.8, 14.5	0.20	+0.5	−0.2, +1.2	0.17
Median	21.7	21.4, 22.1	21.6	21.2, 22.0	0.26	+0.3	−0.2, +0.8	0.25
95th percentile	31.4	30.8, 32.0	31.7	31.1, 32.3	0.06	−0.5	−1.1, +0.1	0.07
FWHM	10.4	10.0, 10.8	10.3	9.9, 10.6	0.52	+0.2	−1.1, +1.5	0.79
Final MRI
Fifth percentile	12.8	12.5, 13.2	13.3	13.0, 13.7	<0.001	−2.0	−2.9, −1.1	<0.001
Median	21.1	20.6, 21.5	21.0	20.6, 21.4	0.70	+0.1	−0.5, +0.7	0.77
95th percentile	32.2	31.5, 32.9	31.4	30.7, 32.0	0.004	+1.2	+0.4, +2.0	0.003
FWHM	12.0	11.4, 12.6	10.6	10.2, 11.0	<0.001	+5.4	+3.4, +7.3	<0.001

R2* values are in s⁻¹. P-values < 0.05 are in bold. FWHM = full-width at half-maximum.

At final MRI, the median R2* within the entire thalamus was still identical ipsilateral and contralateral to infarct. However, ipsilateral to infarct, R2* showed significantly lower fifth percentile (P < 0.001), higher 95th percentile (P = 0.004) and higher full-width at half-maximum (P < 0.001) (Table 1). The asymmetry indices provided consistent results.

These results suggest a spatially heterogeneous distribution of R2* ipsilateral to infarct at 1 year with focal modifications that are poorly represented by median values but better captured using various metrics of the histogram dispersion. Especially, the highest values (95th percentile) might represent iron accumulation, our main variable of interest, which is focally located.

Visual inspection of the average R2* map (obtained from infarcts all flipped onto the right side) at final MRI confirmed a heterogeneous distribution within the thalamus. Especially, the voxel-based comparison revealed significantly higher R2* values mainly observed within the medio-dorsal and pulvinar nuclei ipsilateral to infarct as the major result. A less prominent asymmetry was also observed within the lateral nuclei (Fig. 1).

Figure 1

Spatial heterogeneous distribution of R2* ipsilateral to infarct at 1 year. (A) Average map of R2* within the thalami at final MRI and (B) enlarged view of the average R2* map. The colour range indicates R2* values. (C) Voxel-based comparison of R2* values between ipsilateral and contralateral thalami. The colour range indicates t-values resulting from paired sample t-test after family-wise error rate correction to ensure a false positive rate of P < 0.05 (corresponding to a threshold for t-value of 4.1). (D) Atlas from the thalamic nuclei projected onto a white-matter-nulled MPRAGE template within the MNI152 space. Significantly higher R2* values are visualized within the medio-dorsal nucleus (black arrow, blue on the atlas) and within the pulvinar nucleus (black arrowhead, yellow on the atlas) in infarct side compared to the contralateral side. Foci of increased R2* are also observed within the lateral group (white arrow, orange on the atlas).

Qualitative review of all the individual cases confirmed obvious area of high R2* values focally located within these nuclei (medio-dorsal, pulvinar, lateral group nuclei) ipsilateral to infarct in 25 patients (15%) at final MRI (Fig. 2).

Figure 2

Illustrative cases of focal area of high R2* values within the medio-dorsal nucleus (Cases 1 and 2), the pulvinar nucleus (Cases 3 and 4) and the lateral group (Case 5). Case 1: A 41-year-old female with a right middle cerebral artery infarct, involving the frontal, parietal and insular cortices and the deep nuclei at initial DWI. The initial R2* map showed homogeneous and symmetric thalami. A focal area of high R2* values in the medio-dorsal nucleus ipsilateral to infarct appeared on the final R2* map (arrow) and on the enlarged view of the final R2* map. This focal iron accumulation was also well identified on the final raw T₂* sequence with increasing echo time values (8.7 and 26.1 ms) (arrows). Quantitative R2* measurements within the thalami demonstrated a very high final AI₉₅ measured at +12.4 in the entire thalamus and +33.0 in the medio-dorsal nucleus. Case 2 shows another illustrative example with the same pattern from a 54-year-old male with a left middle cerebral artery infarct, involving the deep nuclei at initial DWI, sparing the cortex. Case 3: A 86-year-old male with a right middle cerebral artery infarct, involving the parietal, temporal and insular cortices and the deep nuclei at initial DWI. The initial R2* map showed homogeneous and symmetric thalami. A focal area of high R2* values in the pulvinar nucleus ipsilateral to infarct appeared on the final R2* map (arrow) and on the enlarged view of the final R2* map. This focal iron accumulation was also well identified on the native T₂* sequence with increasing echo time values (8.7 and 26.1 ms) (arrows). Quantitative R2* measurements within the thalami demonstrated an increased final AI₉₅, measured at +7.3 in the entire thalamus and +2.6 in the pulvinar nucleus. Case 4 shows another illustrative example with the same pattern from a 46-year-old man with a left posterior cerebral artery infarct, involving the occipital cortex at initial DWI. Case 5: A 46-year-old male with a right middle cerebral artery infarct, involving the deep nuclei at initial DWI, sparing the cortex. The initial R2* map showed homogeneous and symmetric thalami. A focal area of high R2* values in the lateral group ipsilateral to infarct appeared on the final R2* map (arrow) and on the enlarged view of the final R2* map. This focal iron accumulation was also well identified on the native T₂* sequence with increasing echo imte values (8.7 and 26.1 ms) (arrows). Quantitative R2* measurements within the thalami demonstrated an increased final AI₉₅, measured at +5.0 in the entire thalamus and +4.4 in the lateral group.

Individual histograms of R2* distributions within the thalamus ipsilateral and contralateral to the infarct lesion from illustrative cases matched also well with the general distribution within the entire population (Fig. 3), i.e. demonstrating a higher 95th percentile within the ipsilateral thalamus.

Figure 3

Histograms and fitted normal curves of the R2* distribution at final MRI within the thalami in the five illustrative cases presented in Fig. 2. In all cases, the R2* values within the thalamus contralateral to infarct (blue) fit well the normal distribution whereas the distribution of R2* values within the ipsilateral thalamus (red) is skewed to the right with higher 95th percentile (arrows). Of note, low R2* values were also more frequent within the ipsilateral thalamus in Cases 1 and 3.

Of note, lower values (fifth percentile) within the ipsilateral thalamus were also observed (Table 1) as illustrated in some individual histograms (Fig. 3) and might represent other forms of thalamic alterations responsible for R2 (and thus R2*) decrease such as oedema, gliosis or demyelination (Supplementary Fig. 4).

AI₉₅ within thalamic nuclei is strongly linked to the initial infarct location in relation to thalamic connectivity

Three patterns were observed from the individual cases, which were consistent with the final average R2* map from the entire population. Focal area of high R2* values within the medio-dorsal nucleus was observed in 13 patients all with more anterior infarct lesions (involving the frontal cortex and the lenticular nucleus) (Fig. 2, Cases 1 and 2). Focal area of high R2* values within the pulvinar nucleus was observed in 11 patients all with more posterior infarct lesions (involving the temporo-parietal and occipital cortices) (Fig. 2, Cases 3 and 4). High R2* values were focally observed within the lateral group nuclei in only one patient with an infarct involving the lenticular nucleus (Fig. 2, Case 5).

To quantitatively demonstrate the relationship between high AI₉₅ and initial infarct location, we first compared final AI₉₅ within the medio-dorsal nucleus (and then pulvinar and lateral group) depending on the initial infarct location. Final AI₉₅ within the medio-dorsal nucleus were indeed significantly higher in patients with infarct involving frontal (P = 0.02), temporal (P = 0.04) and insular cortex (P = 0.03), than in patients without such involvement (Table 2). Conversely, final AI₉₅ within the pulvinar nucleus were significantly higher in patients with infarct involving parietal (P = 0.005) and temporal cortex (P < 0.001), than in patients without such involvement (Table 2). Final AI₉₅ within the lateral group were higher in patients with infarct involving temporal cortex and the lenticular nucleus (Supplementary Table 2).

Table 2

Final asymmetry index of R2*-95th percentile in the medio-dorsal and pulvinar nuclei by anatomic location of infarct (n = 172 patients)

Anatomic location of infarct	Final AI95 within the medio-dorsal nucleus			Final AI95 within the pulvinar nucleus
	Involved	Not involved	P-value	Involved	Not involved	P-value
Frontal cortex	+3.7 (+0.9, +6.5)	−0.6 (−3.2, +1.9)	0.02	+0.6 (−0.9, +2.1)	+1.1 (−0.2, +2.5)	0.62
Parietal cortex	+2.8 (+0.2, +5.5)	+0.1 (−2.6, +2.8)	0.16	+2.4 (+0.8, +3.9)	−0.4 (−1.7, +0.8)	0.005
Temporal cortex	+4.6 (+0.9, +8.3)	+0.2 (−2.0, +2.4)	0.04	+5.1 (+3.0, +7.2)	−0.7 (−1.7, +0.4)	<0.001
Occipital cortex	−0.3 (−6.6, +6.0)	+1.6 (−0.4, +3.6)	0.53	+2.3 (−0.2, +4.8)	+0.7 (−0.4, +1.8)	0.29
Insular cortex	+3.6 (+1.0, +6.3)	−0.5 (−3.1, +2.2)	0.03	+1.3 (−0.3, +2.9)	+0.5 (−0.7, +1.8)	0.44
Caudate nucleus	+2.8 (−0.8, +6.4)	+0.7 (−1.6, +2.9)	0.30	+0.2 (−1.7, +2.1)	+1.2 (+0.0, +2.4)	0.35
Lenticular nucleus	+2.0 (−1.6, +5.7)	+1.0 (−1.2, +3.3)	0.62	+0.3 (−1.6, +2.2)	+1.2 (+0.0, +2.3)	0.42

Asymmetry indices are presented as mean (95% CI). P-values < 0.05 are in bold. AI₉₅ = asymmetry index of R2*-95th percentile.

To validate the strong relationship between high AI₉₅ in a specific nucleus and initial infarct location at the voxel level, we used the VLSM method. The VLSM maps confirmed that infarcts involving anterior voxels in the frontal and temporal lobes and the deep nuclei were significantly associated with higher final AI₉₅ within the medio-dorsal nucleus whereas infarcts involving more posterior voxels at the temporo-parietal junction were significantly associated with higher final AI₉₅ within the pulvinar nucleus (Fig. 4). Infarcts involving the postcentral, precentral cortex and the lenticular nucleus were significantly associated with higher final AI₉₅ within the lateral group (Supplementary Fig. 5).

Figure 4

VLSM to quantify the impact of infarct location on final AI₉₅ within the medio-dorsal nucleus (A and B) and within the pulvinar nucleus (C and D) at 1 year post-stroke. The colour range indicates Z-scores resulting from Brunner–Menzel testing and is overlaid on a 3D T₁-weighted image registered to the standard MNI152 space atlas. Lower Z-scores (red) indicate brain regions associated with higher AI₉₅. (A) VLSM map for infarct voxels associated with higher AI₉₅ in the medio-dorsal nucleus corrected for multiple comparisons after false discovery rate at P = 0.05 resulting in a threshold for Z-score of −2.50 and (B) at P = 0.01 resulting in a threshold of −4.05. Anterior voxels in the frontal lobe (arrow) and the lenticular nucleus (arrow head) remained associated with higher AI₉₅ in the medio-dorsal nucleus at P = 0.01. (C) VLSM map for infarct voxels associated with iron accumulation in the pulvinar nucleus corrected for multiple comparisons after false discovery rate at P = 0.05 resulting in a threshold for Z-score of −2.23 and (D) at P = 0.01 resulting in a threshold of −3.19. Posterior voxels in the temporal and parietal lobes remained associated with higher AI₉₅ in the pulvinar nucleus at P = 0.01. (E) Projection of the infarct voxels significantly associated with higher AI₉₅ in the medio-dorsal nucleus (blue) and the pulvinar (yellow) at P = 0.05 (thresholds from A and C) and P = 0.01 (threshold from B and D) on the 3D cortical surface of a normalized brain. Merged blue and yellow colours appear in green.

To ensure that stroke location was an independent predictor of high AI₉₅ within a specific nucleus unbiased by confounders such as stroke volume, we performed logistic regression analyses with the aim to predict iron accumulation at 1 year within each thalamic nucleus (dichotomized as high or low). Consistently with the previous results, frontal (P = 0.05), temporal (P = 0.02) and lenticular (P = 0.01) locations were independent predictors of high AI₉₅ within the medio-dorsal nucleus in multiple regression after adjustment for age, gender, infarct volume and presence of extra-thalamic haemorrhage (Table 3). Parietal (P = 0.046) and temporal (P < 0.00l) locations were independent predictors of high AI₉₅ within the pulvinar nucleus in multiple regression (Table 3). Lenticular location was an independent predictor of high AI₉₅ within the lateral group (P < 0.00l) (Supplementary Table 3).

Table 3

Logistic regression analysis: predictors of high AI₉₅ within the medio-dorsal nucleus ipsilateral to infarct (n = 172 patients) and pulvinar nucleus ipsilateral to infarct (n = 172 patients)

	Simple regression		Multiple regression
	OR (95% CI)	P-value	OR (95% CI)	P-value
Predictors of high AI95 within the medio-dorsal nucleus ipsilateral to infarct
Agea	0.97 (0.75, 1.26)	0.81
Male gender	1.13 (0.52, 2.43)	0.76
Infarct volumea	1.10 (1.02, 1.18)	0.02
Extrathalamic haemorrhage	1.72 (0.85, 3.48)	0.13
Frontal infarct	1.77 (0.88, 3.56)	0.11	2.13 (1.00, 4.52)	0.05
Parietal infarct	0.91 (0.46, 1.83)	0.80
Temporal infarct	2.49 (1.19, 5.22)	0.02	2.41 (1.12, 5.17)	0.02
Occipital infarct	0.70 (0.22, 2.22)	0.54
Insular infarct	1.18 (0.59, 2.36)	0.65
Caudate infarct	2.56 (1.25, 5.23)	0.01
Lenticular infarct	2.27 (1.12, 4.62)	0.02	2.70 (1.26, 5.79)	0.01
Predictors of high AI95 within the pulvinar nucleus ipsilateral to infarct
Agea	1.03 (0.80, 1.34)	0.80
Male gender	0.72 (0.35, 1.51)	0.39
Infarct volumea	1.10 (1.02, 1.19)	0.01
Extrathalamic haemorrhage	1.53 (0.76, 3.09)	0.24
Frontal infarct	1.13 (0.57, 2.26)	0.72
Parietal infarct	2.68 (1.31, 5.49)	0.007	2.17 (1.01, 4.63)	0.046
Temporal infarct	5.30 (2.50, 11.22)	<0.001	4.70 (2.19, 10.09)	<0.001
Occipital infarct	1.60 (0.60, 4.26)	0.35
Insular infarct	1.60 (0.80, 3.20)	0.19
Caudate infarct	0.97 (0.46, 2.01)	0.92
Lenticular infarct	0.87 (0.42, 1.81)	0.71

Factors associated with the dependent variable with a P-value < 0.2 in univariate analysis were entered into the logistic backward stepwise multivariate model. P-values < 0.2 in univariate analysis and <0.05 in multivariate analysis are in bold.

^aContinuous variable, odds ratio calculated per 10-unit increase. AI₉₅ = asymmetry index of R2*-95th percentile.

Thalamic AI₉₅ at 1 year is an independent predictor of functional, cognitive and emotional outcome

Functional outcome

The distribution of modified Rankin Scale scores in patients with high AI₉₅ in comparison with the others suggested a shift toward worse prognosis (Supplementary Fig. 6).

Using a sliding dichotomy approach, high AI₉₅ within the thalamus was associated with a poor functional outcome evaluated by modified Rankin Scale [odds ratio (OR) = 2.22; 95% confidence interval (CI) = 1.10–4.49; P = 0.03] in simple regression analysis. The association remained significant in multiple regression analysis (adjusted OR = 2.11; 95%CI = 1.03–4.32; P = 0.04) (Table 4).

Table 4

Association between AI₉₅ measured in the entire thalamus and clinical outcome (n = 172 patients)

	Functional outcome		Cognitive outcome				Emotional outcome
	Modified Rankin Scale		MoCA Score ≤ 25		Isaacs Set Test < 28		HADS-A Score ≥ 8		HADS-D Score ≥ 8
	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P
Simple regression
All patients (n = 172)	2.22 (1.10, 4.49)	0.03	1.74 (0.84, 3.57)	0.13	2.75 (1.23, 6.16)	0.01	2.10 (1.00, 4.43)	0.05	2.21 (0.96, 5.06)	0.06
Left infarcts (n = 87)	2.33 (0.77, 7.05)	0.13	7.27 (1.49, 35.5)	0.01	3.97 (1.16, 13.6)	0.03	2.94 (0.90, 9.60)	0.07	1.18 (0.33, 4.25)	0.80
Right infarcts (n = 85)	2.05 (0.81, 5.18)	0.13	1.06 (0.42, 2.71)	0.90	3.57 (1.01, 12.6)	0.048	1.71 (0.64, 4.58)	0.29	7.85 (1.88, 32.8)	0.005
Multiple regression
All patients (n = 172)	2.11 (1.03, 4.32)a	0.04	1.39 (0.55, 3.50)b	0.49	3.47 (1.16, 10.3)b	0.03	2.23 (1.04, 4.79)a	0.04	2.77 (1.14, 6.72)a	0.02
Left infarcts (n = 87)	1.40 (0.40, 4.98)c	0.60	5.36 (0.97, 29.7)d	0.05	2.02 (0.34, 11.8)d	0.44	3.06 (0.87, 10.8)c	0.08
Right infarcts (n = 85)	1.99 (0.79, 5.05)c	0.15			4.16 (0.98, 17.7)d	0.05			7.85 (1.88, 32.8)c	0.005

Associations with the dependent variable with a P-value < 0.2 in simple regression were tested in a logistic stepwise multiple regression model. P-values < 0.2 in simple regression and <0.05 in multiple regression are in bold.

^aOdds ratio adjusted for age, gender, initial DWI volume and lateralization of infarct.

^bOdds ratio adjusted for age, gender, white matter changes, initial DWI volume, lateralization of infarct and education.

^cOdds ratio adjusted for age, gender and initial DWI volume.

^dOdds ratio adjusted for age, gender, white matter changes, initial DWI volume and education.

Cognitive outcome

High AI₉₅ within the thalamus was independently associated with a poor cognitive outcome evaluated by Isaacs Set Test (adjusted OR = 3.47; 95%CI = 1.16–10.3; P = 0.03) (Table 4).

There was no significant association between AI₉₅ within the thalamus and MoCA score (adjusted OR = 1.39; 95%CI = 0.55–3.50; P = 0.49) on the entire population. However, taking into account the laterality of the infarct lesions, we found a strong and independent association with a poor cognitive outcome evaluated by MoCA score in patients with left hemispheric infarcts (adjusted OR = 5.36; 95%CI = 0.97, 29.7; P = 0.05) (Table 4).

Furthermore, we evaluated the impact of high AI₉₅ at the nucleus level on cognitive domains supposed to be more specifically driven by the pulvinar nucleus (visuospatial skills) and the medio-dorsal nucleus (anterograde memory) by using the MoCA subscores (Supplementary Table 4). AI₉₅ within the pulvinar nucleus was significantly associated with alteration of the visuospatial MoCA subscore (adjusted OR = 3.14; 95%CI = 1.03, 9.57; P = 0.045). AI₉₅ within the medio-dorsal nucleus nevertheless failed to be significantly associated with alteration of the recall MoCA subscore.

Emotional outcome

High AI₉₅ within the thalami was independently associated with post-stroke anxiety defined by HADS-A score (adjusted OR = 2.23; 95% CI = 1.04, 4.79; P = 0.04) and with post-stroke depression defined by HADS-D score (adjusted OR = 2.77; 95% CI = 1.14, 6.72; P = 0.02) (Table 4).

At the nucleus level, we found different topographic patterns: high AI₉₅ within the pulvinar nucleus was very close to predicting post-stroke anxiety in patients with left-hemispheric infarct (adjusted OR = 3.12, 95% CI = 0.96, 10.1; P = 0.058) whereas high AI₉₅ within the medio-dorsal nucleus was an independent predictor of post-stroke depression in patients with right-hemispheric infarct (adjusted OR = 6.07; 95% CI = 1.43, 25.8; P = 0.01) (Supplementary Table 4).

Discussion

In this longitudinal prospective cohort study we measured high R2* values within the thalamus ipsilateral to the infarct after a delay that we interpret as iron accumulation, which actually appears with a focal distribution pattern especially within the medio-dorsal and pulvinar nuclei. The nucleus that is involved is strongly determined by the initial infarct location, in line with the known connectivity between thalamic nuclei and cortical areas or deep nuclei. Such secondary alterations contribute independently to the post-stroke functional, cognitive and emotional outcome.

We used R2* mapping, a non-invasive method sensitive to iron concentration (Langkammer et al., 2010), as a biomarker to identify, characterize and quantify remote thalamic degeneration ipsilateral to extra-thalamic infarct in a clinical setting. Magnetic resonance diffusion (Herve et al., 2005) and magnetic resonance spectroscopy (Li et al., 2011) have also been used to quantify thalamic neurodegeneration but at a spatial resolution that was too low to detect focal changes at the scale of the thalamic nuclei. By using higher resolution R2* mapping, we highlighted the spatial heterogeneity and the focal pattern of remote thalamic degeneration. This first quantification in humans is consistent with histopathological studies in rat models (Justicia et al., 2008; Walberer et al., 2014; Weishaupt et al., 2015), which have shown focal iron accumulation in specific thalamic nuclei, especially the medio-dorsal and pulvinar nuclei. From a radiological point of view, these foci of iron accumulation were obvious by simple visual inspection even on the raw T₂* images in about 15% of the patients and should not be misinterpreted as new haemorrhagic lesions. Noteworthy were the low infarct volume and the moderate severity of our population compared to previous reports (Ogawa et al., 1997; Herve et al., 2005), which further support the sensitivity of this R2* approach.

Our results provide additional information concerning the temporality of remote neurodegeneration after infarct. Van Etten et al. (2015) reported a lower average T2*-signal intensity in manually drawn region of interest within the entire thalamus at a single slice level as early as Day 1 after stroke onset. However, T2* signal intensity suffers from high intra- and inter-examination variability and we believe our quantitative R2* mapping method is more robust, particularly with the unbiased 3D atlas-driven regions of interest that we used here. Taken together, our results and those from others (Herve et al., 2005; Justicia et al., 2008; Li et al., 2011), indicate that remote thalamic alterations require time to develop, and are a secondary dynamic process that continues during the chronic post-stoke period. Remote neurodegeneration has been well studied in animal models (Viscomi and Molinari, 2014), and implies a series of effectors at specific time points such as mitochondrial dysfunction, neuronal death, microglial activation/migration and oxidative stress. The iron accumulation observed in vivo here may be mainly mediated by iron-loaded microglial activation and/or extracellular iron around plaque-like amyloid deposits. In a rat model of transient middle cerebral artery occlusion, Justicia et al. (2008) demonstrated a co-location of microglial cell expressing heme-oxygenase HO-1 (a chaperone protein induced under oxidative stress, degrading heme and generating iron) and focal iron accumulation in the ipsilateral thalamus after several weeks to months. PET imaging studies evaluating the time course of microglial activation after ischaemic stroke have also reported progressive changes in the distribution of ¹¹C-(R)-PK11195 signal with delayed involvement of the ipsilateral thalamus (Pappata et al., 2000; Gerhard et al., 2005).

To demonstrate that focal iron accumulation was directly linked to the initial infarct location we combined different approaches, all converging to the same strong finding. Measurements of R2* within the medio-dorsal, pulvinar and lateral group nuclei were significantly different according to the initial infarct location; a result that we confirmed at the voxel level with maps derived from the VLSM method. Logistic regression aiming at predicting focal iron accumulation (dichotomized as high or low) also showed consistent results and ensured that the association between focal iron accumulation within the medio-dorsal nucleus (and pulvinar or lateral group) and the initial infarction was independent of cofounders. In agreement with histopathological (Mengual et al., 1999; Mitchell and Chakraborty, 2013) and imaging connectivity studies of cortico-thalamic connections (Johansen-Berg et al., 2005; Buchsbaum et al., 2006; Leh et al., 2008; Rosenberg et al., 2009), focal iron accumulation within the medio-dorsal nucleus was observed in more anterior infarct lesions involving the frontal cortex and the anterior part of the temporal cortex; focal iron accumulation within the pulvinar nucleus was associated with more posterior infarct lesions involving the parietal, temporal and occipital cortices; and focal iron accumulation within the lateral group was associated with infarct lesion involving the somato-sensory cortex. Interestingly, while most previous studies have been focused on cortical infarcts (Ogawa et al., 1997; Herve et al., 2005), we also observed that lenticular infarcts were significantly associated with iron accumulation within the medio-dorsal and the lateral group nuclei. The lateral group and the medio-dorsal nucleus are major sites of striatal and pallidal projections (Mengual et al., 1999; Mitchell and Chakraborty, 2013). This finding is a unique illustration, to our knowledge, of the concept of thalamo-striatal degeneration after infarct (Behrens et al., 2003; Johansen-Berg et al., 2005).

The thalamus contributes to and regulates a wide range of cognitive processes (Saalmann and Kastner, 2015). We highlighted the clinical relevance of remote thalamic degeneration on post-stroke cognitive impairment evaluated with Isaacs Set Test. This fluency test is sensitive to subtle changes and is reliable to evaluate both normal and severely impaired patients (Proust-Lima et al., 2007; Mehrabian et al., 2015). Consistent with these results, by exploring remote thalamic microstructural abnormalities with diffusion tensor imaging in 17 patients at 3 months post-stroke, Fernandez-Andujar et al. (2014) reported an association between decrease fractional anisotropy in the entire thalami and lower verbal fluency performance. Using MoCA, remote thalamic degeneration was associated with cognitive impairment only in the subgroup of patients with left-hemispheric infarct, which is consistent with previous data indicating the major contribution of the left hemisphere to MoCA performances even independently of language deficits (Munsch et al., 2016).

Given the specific function of each thalamic nucleus, different symptoms could be expected according to the location of the neurodegenerative thalamic lesions. Lesion studies in animal models and patients indicate a specific role of the medio-dorsal nucleus in anterograde memory (Mitchell and Chakraborty, 2013), whereas the pulvinar nucleus is implicated in visuospatial skills and social cognition (Benarroch, 2015). Even though our results suggest such a contribution of remote thalamic lesions at the nucleus level on specific cognitive impairment, they did not reach significance probably because of a lack of statistical power regarding the impact of infarct itself. Furthermore, MoCA provides an overall quantification of post-stroke cognitive disability but MoCA subscores might not be sensitive enough to analyse individual cognitive domains.

We also emphasized the independent impact of remote thalamic degeneration on emotional behaviour, which has never been reported before but which is consistent with recent models on emotional processing (Arend et al., 2015) and with previous studies evaluating patient with direct thalamic infarcts (Santos et al., 2009; De Witte et al., 2011). In agreement with Liebermann et al. (2013) who evaluated the affective status of 68 patients following thalamic infarcts, our results suggested a specific involvement of secondary alterations of the pulvinar nucleus in the development of post-stroke anxiety, which could be mediated by projections to the amygdala (Jones and Burton, 1976), a central actor in the fear-perception circuit involved in anxiety disorders (Stahl, 2003). In particular, a magnetoencephalographic study demonstrated that the pulvinar-amygdala connection enables a rapid treatment of the emotional valence during face processing (Garvert et al., 2014). On the contrary, post-stroke depression was associated with secondary alterations of the medio-dorsal nucleus in our study, independently of stroke location. Interestingly, the magnocellular part of the medio-dorsal nucleus is involved in two frontal-subcortical circuits regulating emotion: the orbitofrontal circuit, which modulates empathy and socially appropriate behaviour and the anterior cingulate circuit, which produces motivation (Burruss et al., 2000). Supporting this model, modification of the number of neurons within the medio-dorsal nucleus (Young et al., 2004), decreased functional connectivity between the medio-dorsal nucleus and anterior cingulate cortex (Anand et al., 2009) and association between elevated microglial density within the medio-dorsal nucleus and suicide (Steiner et al., 2008) have been reported in patients with major depression.

Our study is not without limitations. First, we acknowledge that, while efforts have been made not to select the patients, our sample is constituted of non-consecutive patients, characterized by low infarct volume and moderate severity, and it might not be fully representative of the general population with supratentorial infarcts. Then, we focused our investigations on the thalamus because of its central involvement in many cognitive processes and its connections with widespread cortical areas. We cannot exclude that loss of connectivity with other structures could have also influenced the clinical outcome, independently from the thalamic alterations studied here. The thalamic atlas-based approach allowed us to measure the R2* variations in nuclei, which are very challenging to delineate on standard magnetic resonance sequences. Nevertheless, interindividual differences of nuclear shape might have influenced the regional R2* measurements, even if the co-registration of the atlas to the anatomy of each individual should enable accurate measurements. A broader neuropsychological battery should furthermore be used in future studies to understand more precisely the real impact of the secondary alterations focally located within specific thalamic nuclei that we have not been able to fully explore with our brief sample of cognitive functions. Finally, even if the amplitude of R2*differences between the infarct side and the contralateral side were small, they may be biologically relevant as slight dysregulations of cellular iron concentration are sufficient to trigger a cascade of events leading to cellular death via the Fenton chemistry and oxidative stress or via ferroptosis, a specific iron-dependent cell death pathway (Belaidi and Bush, 2016).

To conclude, R2* mapping overlaid onto an accurate thalamic atlas is a suitable method to demonstrate and quantify secondary remote neurodegeneration after infarct at the scale of the thalamic nucleus, in relation to the pattern of connectivity between thalamic nuclei and cortical areas or deep nuclei. These secondary lesions have the potential to independently impact long-term functional, cognitive and emotional outcome, although larger longitudinal studies will be needed to confirm this statement. This secondary neurodegenerative stream, whose final consequence can be quantified in vivo with R2* mapping, could be a potential therapeutic target for neuronal protection after ischaemic stroke.

Abbreviations

Abbreviations
 
• AI₉₅

  asymmetry index of the 95th percentile of R2*

 
• DWI

  diffusion-weighted imaging

 
• HADS

  Hospital Anxiety and Depression Scale

 
• MoCA

  Montreal Cognitive Assessment

 
• VLSM

  voxel-based lesion-symptom mapping

Acknowledgements

The authors thank Dr Paul Perez and Dr Julien Asselineau (Pôle de santé publique, Unité de Soutien Méthodologique à la Recherche Clinique et Epidémiologique, CHU de Bordeaux, Bordeaux, France) for critical reading and suggestions with the preparation of the manuscript.

Funding

The study was supported by public grants from the French Agence Nationale de la Recherche within the context of the Investments for the Future Program, referenced ANR-10-LABX-57 and named “TRAIL” (Translational Research and Advanced Imaging Laboratory). The study was funded by a public grant from the French government (PHRC protocole hospitalier de recherche clinique inter-régional) funded in 2012. T.T. also received financial support from ARSEP foundation and from French Agence Nationale de la Recherche within the context of the Investments for the Future program referenced ANR-10-LABX-43 named BRAIN.

Supplementary material

Supplementary material is available at Brain online.

Author notes

See Duering and Schmidt (doi:10.1093/awx135) for a scientific commentary on this article.