Introduction

The basal ganglia are composed of five principal deep brain structures: the caudate nucleus and putamen (which comprise the striatum), globus pallidus, subthalamic nucleus, and substantia nigra (the latter located in the midbrain). This set of nuclei is intertwined in a broader basal ganglia-thalamocortical network that takes part in motor, cognitive, and limbic functions. The striatum is the principal input structure of the basal ganglia and receives afferents from the cerebral cortex, thalamus, and brain stem. After a series of internally reciprocal projections, the basal ganglia transmit their output to the thalamic nuclei that, in turn, project this output back to the same areas of the frontal cortex from which the input originated, thus closing a fronto-basal ganglia-thalamic-frontal loop [1].

The debate on the role of subcortical structures in language processing has emerged in the second half of the nineteenth century, but it is yet far from a definitive closure. Some scholars [2] claim that subcortical structures play no specific role in language processing except for the thalamus. Other authors [3, 4] defend that “subcortical aphasia” exists as a unique clinical entity distinguishable from the classical cortical aphasias, although most researchers agree that its pathophysiology and neuropsychology are not well understood. Moreover, the intrinsic complexity of subcortical circuits has led to a dichotomization of subcortical aphasias as “thalamic” and “non-thalamic” (the latter referring to aphasias caused by lesions in the basal ganglia and subcortical white matter pathways).

As early as 1872, Broadbent [5] proposed that the basal ganglia “generated” words in the same way that it did with motor acts, but scholars did not fully embrace this idea. Kussmaul in 1877 [6] attributed a purely motor role to the basal ganglia while Carl Wernicke, in 1874 [7] and Lichtheim, in 1885 [8], defended the notion that subcortical lesions led to language impairment only by disruption of the connecting pathways among the cortical language areas, thus excluding any direct involvement of the basal ganglia and thalamus in higher mental functions.

One of the earliest descriptions of language alterations caused by subcortical lesions was made by Pierre Marie in 1906 [9], who labeled as “anarthria” the speech impairment caused by lesions in the caudate nucleus, putamen, internal capsule, and thalamus. Later in the twentieth century, studies conducted in patients with Parkinson’s disease submitted to stereotactic surgery led to the observation that either the lesion or stimulation of the thalamus and globus pallidus provoked language alterations during and after surgery. Thus, in 1959, Penfield and Roberts [10] suggested that the thalamus played an integrative role in language. A few years later, Schuell et al. [11] stated that the thalamus was involved in the preverbal feedback concerning the adequacy of formulated responses.

However, significant advances in the study of subcortical lesions followed the advent of neuroimaging techniques, which allowed the precise location of subcortical lesions in larger cohorts of subjects with aphasia. Computerized tomography (CT) and magnetic resonance imaging (MRI) have shed light on the relationship between deep cerebral lesions and aphasic symptoms. Functional neuroimaging studies, such as single-photon emission computerized tomography (SPECT), positron emission tomography (PET), and functional MRI (fMRI), and electrophysiology-based methods (cortical evoked potential), further expanded our knowledge about the role of subcortical structures in language processing. In this paper, we review the main anatomo-clinical aspects regarding subcortical aphasias following stroke.

Non-thalamic Aphasias

Non-thalamic aphasia is by far the most controversial subtype of subcortical aphasia. There are several complications when it comes to studying anatomo-clinical correlations involving an ensemble of structures that are so tightly and reciprocally connected. For this reason, there is little consensus regarding the clinical findings and pathophysiologic mechanisms proposed to explain the role of non-thalamic subcortical structures in aphasias. We begin by describing the heterogeneous clinical findings following basal ganglia lesions reported in the literature. Subsequently, the four main theoretical frameworks that attempt to address all clinical and neuroimaging findings in language disturbances following basal ganglia lesions are discussed.

Clinical Findings in Non-thalamic Aphasias

In a recent comprehensive review [12], we tried to track as precisely as possible the correlation between the site of lesion and clinical findings in cases of pure basal ganglia lesions published in a range of 50 years (1966–2016). Our review encompassed 303 patients and disclosed two main problems in establishing anatomo-clinical correlations regarding vascular non-thalamic aphasias: (a) rarely one can find isolated lesions impinging over a single structure due to intrinsic characteristic of basal ganglia vascularization; (b) basal ganglia structures are highly interconnected in a network involving reciprocal inhibition and stimulation; thus the effects of a given lesion in one structure may lead to multiple (and sometimes opposite) effects in neighboring connected structures. Therefore, finding specific clinical patterns in basal ganglia lesions is a challenging task. However, there are some consistencies as displayed in Table 1. In about one-third of patients, no language alteration was found regardless of lesion site, especially in isolated lesions of the caudate nucleus (56.2%). Naming deficits occurred in almost half of the patients and were particularly prominent in lesions of the putamen (60.4%) and lentiform nucleus (59.3%). Comprehension and repetition deficits also appeared in nearly one-third of patients, particularly in lenticular and striatocapsular lesions. The highest incidence of language disturbances was found in striatopallidal lesions (75%), but the total sample is too small (4 patients) to draw inferences about this particular damage locus. For a more comprehensive description of the lesion-aphasia correlation profile, please refer to Table 3 in the Appendix.

Table 1 Language alterations found in 180 patients following left basal ganglia lesion (acute stage—less than one month from ictus)
Full size table

Theoretical Frameworks in Non-thalamic Aphasias

  1. 1)

    Disconnection syndrome: according to this proposal, basal ganglia lesions would disrupt language processing by producing a disconnection between Broca’s and Wernicke’s areas. However, this approach implies a role of periventricular white matter (PVWM) rather than the basal ganglia per se in aphasic symptoms. Thus, damage in pathways connecting the medial geniculate body, temporal lobe (in the temporal isthmus), and the callosal auditory association pathways in the posterior PVWM might account for comprehension deficits. Damage to the external capsule, extreme capsule, and arcuate fasciculus would be critical in repetition disturbances and phonemic paraphasias. Lesions in the anterosuperior PVWM would lead to transcortical motor aphasia (TCMA) by disrupting pathways linking Broca’s area to the supplementary motor cortex. Speech disorders can be found in lesions of the superior PVWM and genu of the internal capsule, where descendent corticobulbar fibers run [25]. Naeser et al. [65, 66] also emphasized the role of the PVWM as a major factor leading to language alterations by describing lesions in the putamen and internal capsule that impaired articulation and naming (in anterosuperior PVWM lesions), comprehension deficits and paraphasias (in posterior PVWM lesions), and global aphasia (associated with more extensive anterior, posterior, and superior PVWM lesions).

  1. 2)

    Direct role of basal ganglia in language processing: after the description of nine patients exhibiting non-fluent aphasia with dysarthria and dysprosodia following isolated striatocapsular infarcts, Damasio et al. [3] postulated that aphasia symptoms might result from deficits in movement programming and perception organization performed by the striatum; nevertheless, the authors also recognized that the interruption of white matter pathways connecting the basal ganglia to prefrontal, frontal, and temporal cortices due to internal capsule damage would play a pivotal role in the language disturbances. Later, Cappa et al. [4] and Mega and Alexander [41] proposed that compromised frontostriatal systems would underlie aphasia symptoms such as non-fluency and anomia by disrupting lexical selection mechanisms. Finally, according to Crosson [67, 68], the role of the basal ganglia in semantically induced lexical selection would be that of engaging or disengaging thalamic modulation over the language cortex through caudate-pallidal-thalamic interactions, which might over-inhibit (caudate nucleus lesions) or disinhibit (globus pallidus lesions) the thalamus. Wallesch and Papagno [69] also postulated a monitoring role of the basal ganglia in lexical selection. Activation of the striatum and thalamus (along with parietal and frontal areas) was documented by fMRI in an ambiguity-resolution task, thus reinforcing the hypothesis that the basal ganglia participate in semantic verification and motor programming preselected language segments [70]. A series of studies by Copland et al. [71,72,73] have demonstrated the role of the basal ganglia in the resolution of lexical ambiguities through selective attentional engagement subserved by fronto-subcortical systems.

  1. 3)

    Dysfunction of remote areas by subcortical deafferentation (diaschisis): the well-known existence of functional impairment in intact regions of the brain that connect to a primary lesion site, or diaschisis, led to the notion that cortical lesions are still ultimately responsible for language impairment, even when the primary lesion is circumscribed to subcortical structures. Studies using SPECT have demonstrated that the long-term persistence of neuropsychological impairment (such as neglect and aphasia) is consistently related to the degree of cortical hypoperfusion associated with subcortical lesions, even in the absence of identifiable structural lesions in the cortex [27, 29].

  1. 4)

    Cortical hypoperfusion: a new perspective to explain subcortical aphasia secondary to basal ganglia and PVWM damage emerged from studies investigating cerebrovascular disease and cognitive impairment. Usually, ischemic striatocapsular lesions are caused by occlusion in the proximal segment of the middle cerebral artery (MCA), or less frequently, occlusion of the internal carotid artery (ICA). Therefore, the amount of cortical neuronal loss consequent to the occurrence of an infarct in these territories is determined by the promptness of flow restoration to the affected area, either by MCA recanalization or an efficient anastomotic circulation. Thus, circulatory dynamics plays a crucial role in the development and recovery from aphasia following basal ganglia injury [74]. Nadeau and Crosson [2] have further developed this theory by considering that (a) the MCA is responsible for the irrigation of the perisylvian cortex; (b) the basal ganglia receive blood supply from lenticulostriate branches, which are terminal branches. It follows that MCA or ICA obstruction is prone to cause greater ischemic damage in lenticulostriate territories while cortical areas may be spared due to an adequate anastomotic circulation, which prevents the occurrence of massive cortical infarction. However, a transient neuronal dysfunction may occur, inducing aphasic symptoms, albeit without evidence of cortical structural alterations. As for cases involving the posterior limb of the internal capsule (supplied by the anterior choroidal artery), aphasic symptoms resemble those of a “thalamic aphasia” due to a thalamic-temporal disconnection. Hemorrhagic strokes, in turn, might cause transient neuronal dysfunction (with consequent language impairment) through the pressure exerted by the clot in the adjacent cortex.

Degenerative diseases such as Parkinson’s and Huntington’s disease, which affect the nigrostriatal system and caudate nucleus, respectively, are alternate models for the role of basal ganglia in language processing. Indeed, impairment in verbal fluency, naming, and syntactic processing were described in both conditions [75, 76]. However, it should be pointed that the degenerative process is not restricted to the basal ganglia but also spreads to the cerebral cortex, especially the frontal lobes [77, 78]

The basal ganglia belong to a network involved in executive functions and acquisition of behaviors both in motor and high-order functions, including cognitive processes and empathic and socially appropriate behaviors [1]. Therefore, it is not surprising that speech planning [79] and prosodic processing [80] may be affected by disorders in the basal ganglia. Moreover, recent theories implicate the striatum in sequential and computational aspects of language processing, combinatorial rule application, and procedural learning, like those associated with morphology and syntax [81,82,83]

Considering the complexity of the subcortical circuitry, we believe that a new promising approach to the problem of subcortical aphasia may rely on connectivity studies. Xu et al. [84•], targeting Broca’s and Wernicke’s areas as regions of interest, described that increase in intrahemispheric and decrease in interhemispheric functional connectivity were associated with better recovery from basal ganglia aphasia. They speculate that this finding may be related to a possible role of cortico-subcortical circuits in favoring intrahemispheric (compensatory) mechanisms that contribute to language recovery.

Thalamic Aphasia

The thalamus is so central to language that many authors stand by the idea that it is the only subcortical center for linguistic processing [67]. Notwithstanding, this structure is subdivided into several different nuclei, and not all of them are thought to provoke aphasia. Primarily, focal lesions that consistently engender language symptoms appear to be located to the pulvinar/lateral posterior thalamus, the anterior thalamic nuclei (ATN), and the ventral anterior nucleus (VA) [85,86,87] (Table 2).

Table 2 Thalamic lesion studies published in the last 40 years
Full size table

In order to cover each of these nuclei, we introduce the overall semiological fingerprint of thalamic aphasia: De Witte et al. [94] reviewed a sample of 465 thalamic infarction patients and described their foremost traits: out of “1) fluent output, 2) normal or mildly impaired comprehension skills, 3) normal or mildly impaired repetition, 4) moderate to severe anomia characterized by semantic paraphasias, neologisms, and perseverations, 5) hypophonia and/or mild articulation deficits and 6) reduction of spontaneous speech or verbal aspontaneity”, 63.6% of left infarction cases conformed to at least four of these criteria. In full accordance, aphasia resulting from thalamic insult is best recognized for its fluent, paraphasic production, which may degenerate into jargon, with perseverations and preserved repetition skills [56, 67, 85, 107]. Although the non-fluent language is not as prevalent, it is important to note that word-finding is harshly affected, and spontaneous speech often plummets [56, 85, 94].

Furthermore, one must also point out that these deficits are mostly lexical-semantic in nature. Nevertheless, each nucleus portrays a slightly different profile in regard to these symptoms.

The Pulvinar

The pulvinar’s partaking in language is conspicuous and most unsurprising, seeing that it projects eminently to temporoparietal cortices as well as Broca’s area [109]. For instance, a recent study reported that resting-state functional connectivity between the left pulvinar, left middle temporal gyrus, and left inferior parietal lobule was significantly correlated with picture naming often brought down a few notches by thalamic infarction [110]. Indeed, some theoretical models on language and thalamic aphasia place great importance on the pulvinar. Hart et al. [111] ascribed to it a leading role in lexical-semantic retrieval, and Nadeau and Crosson [2] did so to thalamic aphasia. Accordingly, lesions to the pulvinar/LP are prone to cause semantic paraphasias and fluent aphasia, as well as naming deficits [56, 92, 93, 104, 112, 113], albeit other aphasic symptoms are also common. However, we should consider that this nucleus’s input supply can also be cut short by compromises to other structures. Thus, through the pulvinar alone, various focal lesions hold the potential to impinge detrimentally on language cortices, thalamic, or otherwise. For example, Nadeau and Crosson [2] also proposed that basal ganglia and/or nucleus reticularis damage should cause subcortical aphasia mainly through the pulvinar, which would gate inputs in turn to the language cortices.

The pulvinar may support fine semantic discrimination similar to that already described in sensory processes: pulvinar damage and deactivations consistently undermine visual discrimination tasks [114]. This structure contributes to eliciting the mismatch negativity for tone discrimination in vivo [115], thus helping to resolve competition between representations. We should then consider that the pulvinar may play a role in paraphasia, neologisms, and jargon through similar disorganization of fine semantic discrimination in the language cortices.

Anterior Nuclei

It grows increasingly evident that damage to the anterior thalamic nuclei (ATN) and/or ventral anterior thalamus (VA) robustly results in a sharp plunge in linguistic output (and perseverations): the lack of spontaneous speech and verbal fluency, anomia, as well as the occasional non-fluent aphasia, are most notably observed in anterior thalamic lesions [85, 87, 88, 96, 98, 100, 102, 103, 105, 116, 117]. Likewise, this seems to match an idea proposed by Cox and Heilman [118] about their thalamic aphasia case study, namely, the patient’s lexical-semantic representations seemed intact, but his ability to spontaneously access them was hampered. Such a concept could be intrinsically linked to what may be referred to as a robust disruption of self-initiated generation of language in anterior thalamic aphasia. Furthermore, this may also manifest in perseverations [100] coupled to word-finding difficulties and hypophonia. The study by Fritsch et al. [87] claimed that “(1) aphasic symptoms after an isolated lesion to the thalamus (ITL) are rare (6/52 patients) and that (2) aphasic symptoms after ILT are strongly associated with isolated left anterior thalamic lesion location”. Hence, not only do ATN lesions impinge on generation in language processes, but they are also some of the most common.

Notably, the ATN are thought to propagate theta activity and mediate information transfer between the frontal and (predominantly medial) temporal lobes [119]. In turn, in controlled processing paradigms, the theta range is tightly associated with semantic distance [120, 121] and lexical-semantic retrieval [122]. Frontal and temporal theta activity was linked to word-finding during fluency in a recent study [123]. Leszczyńsk and Staudigl [124] further speculated that the ATN might prop up memory-guided attention. Such data seem to endorse that ATN insults would cut short the spontaneous initiation of more volitional top-down predictions/verbal search, likely through frontotemporal theta synchronizations.

The Motor Thalamus

The VA is pronouncedly connected with Broca’s area and the pre-supplementary motor area (pre-SMA) [125]. Mounting evidence points to the primary effect of VA lesions being a reduction in speech output and fluency, and the incidence of perseverations. Analogically, perseverations of incorrect words in language tasks could be induced by stimulation of more anterior portions of the VL [67, 126,127,128]. Paraphasia is another very robust feature [86, 87, 129], and hypophonia is a fairly common outcome (which is reminiscent of parkinsonian patients, where VA function is altered) [100, 130].

The idea that word selection and retrieval undergo VA and/or pre-SMA processing is not novel [68, 101, 111, 120]. As stated earlier when we referred to the direct role of basal ganglia in language processing, Crosson et al. [68] described a loop consisting of left pre-SMA-dorsal caudate-VA (the GPi theoretically being suppressed by the striatum) that was ignited during word generation, further conjecturing that this circuit would be responsible for selection processes. A role in selection is also concordant with the VA’s intimate connections with the IFG and pre-SMA, which are thought to perform these functions [109]. Since the frontal aslant tract also bridges both, this may implicate the triad in a circuit for word selection; indeed, recent evidence indicates that word-finding difficulties and language in thalamic stroke were strongly correlated with pre-SMA and IFG abnormalities in SPECT recordings [131]. Furthermore, VA lesions might oppose the initiation and/or suppression of speech/selection [132] through a premotor cortex-thalamic network. This idea seems pertinent given that ATN/VA aphasia is fluent despite the reduction in spontaneous speech and the increase in perseverations [100, 104, 108, 119]. Conversely, an MRI study by Nishio et al. [133] claimed that several structures in the rostral vicinity of the ventral intermediate nucleus (VIM) are important for word processing. Tractography analyses in patients with anterior thalamic infarcts have shown that lexical-semantic deficits were related to the disconnection of language-relevant frontal and temporal cortical areas (the middle frontal gyrus) from ventral anterior and ventral lateral thalamic regions.

In that vein, an interesting finding is that parkinsonian patients take a longer time to disambiguate words, lengthening priming of contextually incongruent meanings behaviourally, and electrophysiologically word selection is delayed [134, 135]. When a representation is ambiguous, one of the meanings is favored by contextual variables and word frequency; the VA-VL may respond accordingly with linear tonic transmissions, biasing word selection in the pre-SMA and IFG [136]. Conversely, when the VA is constantly suppressed in Parkinson’s disease, both weak and stronger meanings drive all-or-none bursting amplified by the VA-VL to outlast their due timing, causing erasure of salience [137] and making selection unviable. It would be interesting to explore whether a similar suppression (“disfacilitation” or hyperpolarization) arising from thalamic insults would hamper word selection in a kindred fashion.

Mediodorsal Nucleus

Finally, the mediodorsal nucleus is not commonly linked to aphasia, but it is noteworthy that damage here handicaps verbal memory and causes some measure of anomia, perseveration, and executive deficits [56, 88, 89, 138].

Deep Brain Stimulation Studies

Deep brain stimulation (DBS), a treatment currently used in many neurological and psychiatric disorders, has opened a window of opportunity for observing in vivo human function of subcortical structures in several motor and cognitive domains [139]. The method uses implanted electrodes that deliver current to the brain and also permits electrophysiological recordings of deep brain structures while the individual is awake and capable of performing motor and cognitive tasks. For this reason, DBS procedures performed on patients with motor disturbances have provided an interesting set of data regarding basal ganglia and thalamic participation in language processing.

Basal Ganglia

Nowadays, most DBS procedures on Parkinson’s disease patients target the subthalamic nucleus (STN) and globus pallidus pars interna (GPi) [140]. Post-operative longitudinal language assessment showed that DBS intervention targeting both structures was associated with impairment in phonemic and semantic verbal fluency in several studies [141,142,143]. Moreover, 35–90-Hz stimulation in the STN was correlated with switching (change from one semantic category or letter to another) during VF semantic and phonemic tasks [144]. Wojtecki et al. [145] found increases in 6–12-Hz local field potential (LFP) activity in the STN during a verbal generation task (especially in the ventromedial STN), which was linked to an enhanced coherence between the STN and frontal areas at 6–7 Hz. In the same vein, Hohlefeld et al. [146] described a correlation between lexical accuracy and cortico-subthalamic coherence at the 14–35-Hz range. These data provide evidence for cortex-basal ganglia synchronization in language processing.

Thalamus

Word selection is vulnerable to VA-VL lesions, especially the more anterior portions; a neighboring area, the VIM, is often targeted for DBS treatment of essential tremor, and its stimulation, through current spread, can encroach on the VA-VL complex [147], thereby modulating the latter’s function. This target is convenient for probing a role in language, seeing that VIM infarct per se was shown not to correlate with language symptoms.

A meta-analysis unraveled that both lesions and stimulation thalamic surgery “produce adverse effects on speech. Left-sided and bilateral procedures are approximately 3-fold more likely to cause speech difficulty” [148]. VIM stimulation slowed down verbal fluency, increasing intra-cluster pauses (especially in phonemic fluency) [149]. Intuitively, intra-cluster choices should suffer more significant interference from neighboring nodes, therefore leaving room only for the most salient alternatives to outstand for selection. Consonantly, recent findings documented that VIM stimulation increases the frequency of lexical items in spontaneous speech [150], suggestive of unavailability of weaker tokens, which also applies to thalamic aphasia [101]. Tiedt et al. [145] attributed their findings to a perturbation of thalamic interareal binding [151], “resulting in a lower network connectivity state, reduced lexical activation spread and, finally, slower word production”. Indeed, a recent study documented that acute ischemic thalamic stroke was linked to cortical dysconnectivity [152].

Another recent study assessing left VA-VL (particularly the VA and VLa) stimulation showed that it suppressed verbal abstraction [152]; the same was not true of the VIM. The authors underscored that the VA-VLa receives its principal inputs from the GP and SNr while projecting extensively to the IFG (which supports verbal abstraction), consequently leading to repercussions for verbal abstraction ability. Similarly, perseveration of incorrect words was also induced by stimulation of the VLa in a few studies [67, 127, 128]. Conversely, the VLp is intertwined with cerebellar circuitry and communicates with the non-cognitive motor cortex. In line with that are classic studies by Ojemann [126, 153]: 60-Hz stimulation of the VL prolonged the duration of oral reading responses with slurring and motor distortions (and some other motor-related findings came out concerning VIM stimulation) [154]. Another example of language increments is found in Pedrosa et al. [155]: 10-Hz stimulation of the VL enhanced phonemic and semantic fluency, whereas 120–150-Hz stimulation impaired them, suggesting short-term synaptic depression with high-frequency stimulation [156]. Paradoxically, anomia ensues from 60-Hz VL stimulation most reliably when electrodes are placed near the intralaminar groups [126, 153]. However, there are findings that 50-Hz stimulation of the centromedian nucleus actually enhances motor speech and semantic retrieval [157, 158].

Conclusions

As a conclusion of this review, we believe that some issues are worth mentioning: firstly, we focused our review on left hemispheric lesions; however, it does not imply that language alterations do not occur in subcortical right hemisphere lesions, in both left and right-handed individuals. On the contrary, there are several reports of language impairment following right basal ganglia and thalamic lesions. The findings are heterogeneous regarding right basal lesions, and aphasia occurred mainly in left-handed patients [12]. In right thalamic lesions, lexico-semantic impairment, as revealed by anomia and paraphasias, can be more consistently found, although in most cases there are no reports of patients’ handedness [94, 95]. Secondly, we must take into account that the classification non-thalamic × thalamic aphasias only makes sense (if any) from the structural/vascular perspective, given that language is organized on a multisynaptic network that involves the neocortex, basal ganglia, thalamus, and also the cerebellum forming a closed-loop circuit [159•]. Consequently, an isolated structural lesion may impact the whole functional system, although with different intensity and patterns, irrespectively of its location. Therefore, it is our opinion that in the future, subcortical aphasia studies will benefit primarily from tractography and cortical connectivity studies.