Six patients (4 men; mean age 60.0 ± 12.0 years) with acute ischemic stroke within the middle cerebral artery (MCA) territory presenting with aphasia were enrolled in this study. The diagnosis of ischemic stroke was based on clinical criteria and on the results of diffusion weighted MRI at admission (figure 1). The site of vessel occlusion was demonstrated using Time-of-Flight MR angiography. Vascular risk factors and carotid ultrasound findings were assessed for each patient by reviewing their clinical charts. Tables 1 and 3 detail the patients 'clinical data and the examination times. Since leukoencephalopathy may affect BOLD signals , we excluded patients whose MRI rated >4 for the "Age Related White Matter Changes" score . Patients with a poor insonation of MCAs at the temporal bone windows were excluded since TCD examinations could not be performed.
The hemodynamic response function (HRF) was estimated in 12 healthy, age-matched control subjects (8 men, mean age of 57 ± 9.4 years). Reference values for hemodynamic parameters were derived from a previously analyzed older adult control population (mean age 63 ± 9 years) .
The local ethics committee approved the study (267/05), and informed consent was obtained from all subjects.
Experimental examinations were carried out at the acute phase (Ex 1; i.e., within four days of stroke occurrence) and the subacute phase (Ex 2; i.e., between five and twelve days after stroke occurrence and at least two days after Ex 1). The examinations included:
Transcranial Doppler (TCD) to evaluate (i) patency and blood flow velocity of middle cerebral arteries (MCAs) and (ii) cerebral autoregulation.
fMRI study to assess BOLD signal HRF in bilateral auditory cortices in response to two language paradigms.
The autoregulation and fMRI sessions were performed within six hours of each other.
Neurological status was assessed with the NIH stroke scale (NIHSS) at admission (Ex 0) and before each measurement (Ex 1 and Ex 2).
Cerebral hemodynamics evaluation
Measurements were performed with subjects in a supine position with slight to moderate elevation of the upper body. CBFV in both MCAs was measured by TCD with 2 MHz transducers attached to a head frame (TC2-64, EME, Germany). Continuous ABP recording was achieved via a servocontrolled finger plethysmograph (Finapres, USA). End-tidal CO2 partial pressure was measured in mm Hg with an infrared capnometer (Normocap Datex, Finland) during nasal expiration. After establishing stable values, a data segment of ten minutes was recorded as patients breathed spontaneously.
In order to grade cerebral autoregulation, we used the previously described correlation coefficient method, which makes use of spontaneously occurring fluctuations in ABP and CBFV . This approach is well established in neurocritical care and has been validated against static autoregulation measurements . It is based on the simple assumption that decreasing cerebral autoregulation leads to an increasing correlation between fluctuations in CBFV and ABP (i.e., CBFV depends increasingly on fluctuations in ABP). To quantify this correlation, mean values of ABP and CBFV raw data were first averaged over three seconds. For 20 of these three-second averages (i.e., for one-minute periods), Pearson's correlation coefficients between the mean ABP and CBFV were calculated. The resulting sets of one-minute correlation coefficients gained from the entire time series were then averaged, yielding the autoregulatory index Mx. Mx increases with decreasing dynamic autoregulatory capacity. According to reference ranges defined in an elder population , Mx ≥ 0.46 corresponds to exhausted cerebral autoregulation.
In both paradigms, stimuli were spoken by a female voice and recorded by Cool Edit software. Stimuli were presented binaurally through MR compatible headphones. Patients were asked to listen carefully and press a button at the end of each stimulus to ensure alert listening. In the present study, we were only interested in auditory cortex activation in response to speech stimuli as compared with background noise. Since our Department of Neurology enrolled stroke patients in different fMRI studies, the six patients included in this study underwent two different fMRI language paradigms. However, both paradigms were event-related experiments that differed only in the number of stimuli and sessions.
Language paradigm 1
In an auditory comprehension task, we presented 30 normal speech stimuli (German sentences, e.g., "Der Pilot fliegt das Flugzeug" [English translation: "The pilot flies the plane"]), 30 stimuli of pseudo speech, which was derived from normal speech stimuli by exchanging phonemes (e.g., "Ren simot plieft mas kugireug", [English translation is not possible]), and 30 stimuli of reversed speech (e.g., "guezgulf sad tgeilf tolip red", [English translation is not possible]). The reversed speech stimuli were using the Cool Edit software. Stimuli duration ranged between 1730 and 2720 ms. Stimuli were presented binaurally in a pseudo randomized order with an inter-stimulus interval that varied between 3000 and 6000 ms. Stimuli were assigned to a single nine minute long session.
Language paradigm 2
In a modified version of paradigm 1, we presented 92 stimuli of normal speech (German sentences, e.g., "Der Pilot fliegt das Flugzeug" [English translation. "The pilot flies the plane"]) and of reversed speech (e.g., "guezgulf sad tgeilf tolip red", [English translation is not possible]). Stimuli were presented binaurally in a pseudo randomized order with an inter-stimulus interval that varied between of 3000 and 6000 ms. Stimuli were assigned to six sessions, resulting in a total scanning time of 21 minutes (for details see ref 7).
MRI data acquisition
Functional and structural MRI data from all subjects were acquired on a 3 T Siemens TIM Trio scanner with a standard head coil.
Diffusion weighted imaging (DWI)
Scans were obtained using a standard in-house stroke DWI sequence (23 slices, matrix 128 × 128 pixel2, voxel size 1.8 × 1.8 × 6 mm3, TR = 3.1 s, TE = 79 ms, flip angle = 90°).
In cases where the sequence specifications differ between paradigms, values for language task 1 and 2 are given in parentheses. A total of 1 × 260 (6 × 115) scans per examination with 36 (32) axial slices covering the whole brain was acquired in interleaved (descending) order using a gradient echo echo-planar (EPI) T2*-sensitive sequence [resolution = 3 × 3 × 3 mm3, TR = 2.19 (1.83) s, TE = 30 (25) ms, flip angle = 75° (70°), matrix = 64 × 64 pixel2]. During reconstruction, scans were corrected for motion and distortion artifacts based on a reference measurement.
A high-resolution T1 anatomical scan was obtained (160 slices, voxel size = 1 × 1 × 1 mm3, TR = 2.2 s, TE = 2.6 ms, FOV = 160 × 240 × 240 mm3) for spatial processing of the fMRI data.
fMRI Data analysis
fMRI data were analyzed with SPM5 (http://www.fil.ion.ucl.ac.uk/spm).
Data were pre-processed using standard routines implemented in SPM5. In both experiments, slices were first corrected for different signals acquisition times by shifting the signal measured in each slice relative to the acquisition of the middle slice. Volumes were then spatially normalized to the Montreal Neurological Institute (MNI) reference brain using non-linear normalization parameters that were estimated during segmentation of the coregistered T1 anatomical scan . All normalized images were then smoothed using an isotropic 9-mm Gaussian kernel to account for inter-subject differences. Data were motion corrected during acquisition using the method introduced by Zaitsev et al. 
Finite impulse response (FIR) analysis
The time course of the hemodynamic BOLD response in both hemispheres was estimated using FIR analyses as implemented in SPM5. Onsets of auditory stimuli were convolved with a set of ten successive basis functions, which resulted in ten temporally aligned regressors for each condition. Each single basis function estimated the size of the BOLD signal for a specific time window of length TR/2. Altogether, the condition-specific sets of the ten basis functions covered a total post stimulus time of 10.95 s and 9.2 s for paradigms 1 and 2, respectively (see Figure 2 and 3). F-contrasts were computed across all ten basis functions. In the peak voxels within both hemispheres, parameter estimates were extracted for each basis function. In single subject analyses on study patients, parameter estimates represent the averaged effect size across stimuli, while in random effects group analyses on controls, parameter estimates represent the averaged contrast estimate across subjects. Since parameter estimates resembled normalized values, comparisons across subjects were valid. For comparison across paradigms, contrast estimates were divided by the number of sessions (i.e. in case of paradigm 2 by factor 6). The time bin with the highest contrast estimate was used as an approximation of the TTP, and the contrast estimate itself reflects the amplitude of the HRF.
Statistical analysis was carried out using SPSS 17.0 software. Nonparametric tests (e.g., Wilcoxon signed-rank test for paired data, Spearman test) were used to compare a patient's data from the two hemispheres, to evaluate their variations over time, and to assess possible correlations. Since the sample size was small, we grouped data from both hemispheres in order to assess the correlation between TTP and Mx index. This was done using Generalized Estimating Equations (GEE) models, which allowed us to account for repeated measurements within subjects (in this case, the measurements from two hemispheres). We reported nominal p values and considered p values of p < 0.05 to be statistically significant. We did not adjust p values for multiple comparison adjustments since they would have significantly reduced the ability to detect interesting correlations within this small sample.