Selection of studies

The computerized Embase, Medline Ovid, Web of Science, Cochrane central and Google scholar databases were searched for articles in January 2017 (and again in September 2017 to detect recently published articles) combining the search terms neurological, neurophysiology, neurobiology, forebrain, brain AND speech, language, verbal, linguistic, reading, writing, literacy, illiteracy, vocabulary, grammar, phonology, dyslexia AND premature, prematurity, preterm, "low birth, birthweight", "small*gestational age". In Fig 1 the flow diagram of the study selection is presented. The search yielded 2083 unique articles. In S1 Text the complete search protocol can be found for all databases. Based on screening of titles and abstracts, 2035 articles were excluded. 49 articles remained and were assessed for eligibility based on the following exclusion criteria: (1) study cases are not school-aged children (6–17 years) born preterm (gestational age (GA) <37 weeks); (2) brain structures not measured with structural MRI (T1- and T2-weighted scans) and/or DTI; (3) language not assessed at the same age as the MRI scan was made; (4) no correlational measure is published between language and brain volume or fractional anisotropy of a brain structure; (5) not published in an English language, peer-reviewed journal; (6) no sufficient study quality according to the Newcastle-Ottawa Quality Assessment Scale for cohort studies. A total of 25 studies had to be excluded based on these criteria, which resulted in 24 studies that were suitable for our data extraction and analysis (23 originally in January 2017, and 1 added in September 2017). Subsequently, one study was excluded since there was high risk of bias [14]; the population and main outcome measure of this study were overlapping with those of one of the other included studies [15]. Of these two studies, we included the most recently published one [15]. The main characteristics of the final 23 included studies are presented in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow diagram of study selection.

https://doi.org/10.1371/journal.pone.0196607.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Study characteristics.

https://doi.org/10.1371/journal.pone.0196607.t001

Quality assessment

Two authors (LWS and JD) independently assessed the methodological quality of the included studies according to the Newcastle-Ottawa Quality Assessment Scale for cohort studies (S1 Table). This scale assesses the quality of cohort studies from the selection of the population, the comparability of the study groups and the ascertainment of outcome of interest. The total rating score ranges from 1 to 9, with 9 being the most favorable. Any disagreement between the two assessors with regard to the total score was resolved by discussion. The quality scores of selected articles are presented in S1 Table. Overall quality was rated from 5 to 8 stars. To improve our reporting the PRISMA checklist was used (S2 Table).

Outcome measures

Data extraction

For each included study we extracted the published correlational measures between language scores and brain volume or fractional anisotropy of a brain structure. Most of the studies reported a Pearson’s correlation coefficient. A few studies also reported Spearman’s rho or stepwise logistic regression analyses as a correlation measure. The correlational measure had to correlate a language score (classified within one of the three categories discussed above) and a brain area that is addressed in at least two different studies. For example, total brain volume, brainstem volume and cerebral spinal fluid (CSF) were all studied only once [29–31], and therefore the respective findings are not presented in the cross table. Besides, these studies explicitly reported that they did not find any significant relations with language.

Oral language

Verbal fluency: Language and executive functioning

Written language: Reading and spelling

Main findings

Our overview of study results in language and brain structure associations in preterm children yielded a complex set of relations, of which some show more consensus than others. We will discuss the most remarkable results.

Perhaps most notable is the lack of any association between structural brain injury and language outcomes. We had expected that preterm children with explicit brain damage would have the most severe language problems. However, in these studies brain damage was scored on a 3- or 4-point scale and naturally, in all studies the group of children with explicit damage was relatively small compared to groups of children with less damage, which makes it hard to prove a correlation with language. This might have influenced the correlations found between language skills and brain damage.

Another remarkable result concerns the cerebellum volume. We studied three language domains (i.e. oral language, verbal fluency and written language) and only very few studies found a significant correlation between any of these domains and the volume of the cerebellum; no correlation at all was reported for verbal fluency. A clear correlation between verbal fluency and cerebellar volume cannot be shown, and seems unlikely for both oral and written language.

The association of the CC volume with language outcomes is more convincing, particularly with regard to oral language skills and verbal fluency. Only one of the included studies did not find a significant correlation with language skills or verbal fluency [31], but this can likely be ascribed to insufficient statistical power, as this was the study with the smallest population of preterm children (N = 25). Overall, an association between oral language and CC volume is likely. However, the relation between CC and written language skills remain inconclusive, since there were only two studies that reported correlational data between these measures and they reported opposing results [27, 34].

Regarding the DTI studies, the most striking result is the repeatedly reported significant correlation between UF and oral language skills. The UF is part of the ventral language pathway and in recent literature it is often associated with semantic language functions. However, there is a lack of evidence for a general role of the UF in language [38]. Our review, though, showed a positive correlation between the UF and oral language skills. An association between language skills and the AF, which is part of the dorsal language stream, is less obvious according to our review results. Unfortunately, few studies included in our review addressed the role of the AF. Still, these tentatively show that the AF is more involved in verbal fluency, whilst the UF is more involved in oral language.

A less convincing result is the correlation between language and areas of WM and GM volume. Many studies did look at WM and GM in relation to one of the language domains, but the results were inconsistent. We propose that these kinds of differences between studies might arise because each addresses a slightly different microstructural area of the brain. When total WM or GM volume is studied, rarely any relation with language is found, while many significant relations are found when studying several microstructures of the brain. For example, Nosarti et al. [21] found a negative correlation between written language skills and GM volume in the temporal lobe, while Scott et al. [23] found a positive correlation in the frontal lobe. Overall we see that GM volume is more often negatively correlated with language skills, while WM volume mostly correlated positively with language skills. This negative association of GM with language corresponds with recent literature, also for example in stuttering literature [39], and has been associated with a cortical developmental phase of dendritic and synaptic pruning in late childhood and adolescence [40–43]. This might mark a shift from relatively diffuse cortical representations of cognitive functions in early childhood toward a more accurate, efficient, and faster processing language system later on.

Oral language skills are more often significantly correlated to preterm brain structures compared to verbal fluency skills or written language skills (see Table 2). Thus, atypical brain development in preterm children seems to affect oral language more obviously than it affects verbal fluency or reading or spelling. This is interesting in the sense that verbal fluency skills are also based on executive functioning, while oral language skills are mostly language proficiency tasks. Apparently, brain structures of preterm children are associated more strongly with language tasks than with executive functioning based language tasks.

Influencing factors

It is plausible that gestational age (GA) is an important influencing factor in the relation between brain and language, where lower GA leads to more atypical brain and language development and a relation between these two parameters would be more obvious. However, the populations of almost all included studies consisted of very preterm children with a gestational age of <33 weeks. One study [30] included a population of late preterm children only (34–36 weeks GA). The authors did not find a correlation with language, which is in accordance with the idea that higher GA leads to less atypical brain and language development. However, this lack of correlation might be a consequence of the fact that only the cerebellum was studied, which in many of the other studies was not correlated with language. Because of these considerations we cannot indicate an effect of GA from our study results.

Another factor that might have been of influence is the MRI methodology used. Overall, studies that used DTI as a MRI measurement reported more significant correlations than studies that used volume measurements only.

A third factor to take into account is sex. Several included studies presented results for boys and girls separately. We analyzed these results to search for similarities in boys and girls, but did not find consensus within these results. Therefore, we cannot draw a general conclusion about the role of sex in the correlation between brain and language.

Strengths and limitations

The strength of this review is that it provides a clear overview of the most comparable studies on the relation between language and brain structure in school-aged preterm children. This is a very complex subject because it covers two crossing parameters, which each in itself is complex and variable. We achieved to keep the most important factors relatively stable, such as age of the population, MRI scanner features, language tests used and population size. Also, our classification into three language domain groups resulted in a structured overview. We hope that this categorization will contribute to the validity of future correlational studies of brain function and language outcomes.

A possible limitation of the study is the risk of publication bias, i.e. studies may have analyzed more regions in the brain than reported in the result section. To partly adjust for this, when a certain region was mentioned in the methods section, but not addressed in the results section, we interpreted this as: no significant correlation found. And then, of course, performed studies that did not find any significant correlations may not have been published at all. Therefore, we highlighted the results presented by at least two study groups. When only one study looked into a certain relation–for example the relation between written language skills and the fasciculi—we did not highlight the outcome in our review results and discussion.

In our systematic review we chose to focus on the most commonly used MRI and DTI methods (structural T1- and T2-weighted sequences MRI and DTI). However, there are already some new models, such as non-tensor-based diffusion imaging analyses (e.g. high angular resolution diffusion imaging (HARDI)) that are very promising. HARDI is a new advanced model, which is an improvement with respect to DTI because it can deal with crossing fibers in voxels. It is successfully used very recently in a study with preterm children at term equivalent age [44]. The authors state that their findings suggest that differences in arcuate fasciculi micro-structure have a significant impact on language development and modulate the first stages of language learning. However, non-tensor-based diffusion imaging analyses were beyond the scope of our systematic review since the method is relatively new. Data are still limited and no studies are published in school-aged preterm children yet.

Implications for further research

For future research we would recommend to relate the brain on a microstructural level to oral language skills in preterm children. We would recommend to use language tests such as the CELF, since this test battery consists of a number of subsets that cover all oral language domains and can be subdivided in subcategories, such as perceptive and expressive language. We recommend to study these oral language subcategories separately and relate these and written language skills to brain structures. With respect to MRI/DTI measurements, longitudinal GM and WM analysis seems to be promising methods, highly relevant to longitudinal research in preterm children and the relation with their cognitive development. We also recommend to use non-tensor-based diffusion imaging analysis, since this new advanced model is an improvement with respect to DTI. Besides, we recommend to use principle component analysis (PCA) as an analytical approach, which might prevent random correlational findings and can actually lead to meaningful associations. PCA is a renowned method with a longstanding tradition which is now again increasingly and successfully being used in neonatal MRI studies to quantify the proportion of shared variance in the measured water diffusion parameters (MD, FA, λ_ax and λ_rad) across the tracts [45]. Lastly, more consistent data collection and data sharing could lead to more and quality-assured knowledge in this research field.

Conclusion

This systematic review gives an overview of the extensive and diverse scientific literature on the associations between MRI brain measures and language outcome in children born preterm. Oral language skills and verbal fluency were shown to be associated with CC volume. Oral language skills are also associated with the UF. Overall, oral language skills are more obviously associated with several microstructural brain areas than are verbal fluency tasks, which are executive functioning based language tasks, and reading and spelling tasks. No associations were found between cerebellar volume and verbal fluency. The relation between oral language and written language with cerebellar development seems weak. The relation between preterm brain injury and language outcome could not be proven in studies that used brain damage scales. This most likely implies that not one single damaged brain area is responsible for atypical language development, but that several brain areas and their connections are essential. For future research we would recommend to study overall brain connectivity in combination with oral language skills, in which good quality management and data sharing will be crucial to enhance our shared knowledge and clinical opportunities.