Article activity feed

  1. Evaluation Summary:

    This paper is of interest to developmental neuroscientists who study the early stages of cortical maturation and specialization, particularly in the context of somatosensory and pain system development. The authors suggest that, relative to the infant touch somatotopic map, the infant nociceptive map is more widespread and poorly localised, consistent with infants' poorly directed pain behaviour. However, there are differences in the the implementation of touch and pain conditions and concerns around the analyses that limit support for this interpretation.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

  2. Reviewer #1 (Public Review):

    This manuscript investigated the localisation of infant cortical maps related to touch and noxious stimulation. Using a sample of 32 infants and cot-side measures of evoked brain activity assessed using fNIRS, they localise activity evoked by a non-noxious touch stimulus and a clinically required skin breaking blood test procedure (heel lance). The authors present results that suggest that the infants touch maps more closely overlap the somatotopic heel region than the nociceptive map, and that the nociceptive map is more poorly localised and extends inappropriately into the hand region. However, the conclusions of this paper are mostly unsupported by the results. The following four issues should be noted when reading the current manuscript.

    1. Differences in regions of activations constituting touch maps and nociceptive maps are never actually statistically tested:

    The core finding of this paper is that the infants' nociceptive maps and touch maps are not aligned, they are different: "The key difference between the two patterns of activation is that the foot touch response is limited to the areas of the S1/M1 associated with the foot, whereas the lance response extends towards other more ventral regions of S1". However, this claim of differences between touch and nociceptive maps is never actually statistically tested, and therefore this core finding is currently not backed by the analysis and results.

    The way the analysis is conducted is as follows: (i) a region of activation for the noxious stimulus is found by comparing evoked activity to pre-stimulus baseline and statistically significant regions identified, (ii) a region of activation for the touch stimulus is found by comparing evoked activity to pre-stimulus baseline and statistically significant regions identified, (iii) these noxious and touch activation maps are overlaid and regions of overlap are identified as well as regions of non-overlap, and it is these regions of non-overlap that are used to conclude that the maps differ. However, these analyses are invalid as the only statistical comparisons made are relative to baseline. The activation evoked by the noxious stimulus is never directly statistically compared to the activation evoked by the touch stimulus i.e. the regions of non-overlap differ in their statistical relationship with baseline, not with their statistical relationship to each other.

    The nature of this analysis flaw is expanded upon in the following eLife paper "Science Forum: Ten common statistical mistakes to watch out for when writing or reviewing a manuscript" (, problem number 2 "Interpreting comparisons between two effects without directly comparing them". This eLife paper also advises pointing to the following paper: And the issue is also detailed here: As outlined in the references provided, the way to demonstrate a difference in the two maps in a valid manner is to directly compare them and identify statistically significant differences in regions of activation.

    This error is major flaw in the study, as the inappropriate analysis underlies the core novelty of the paper. All other statistically significant findings in the paper have been previously identified: somatotopic organisation of infant S1 has been reported by Dall'Orso et al., 2018; much larger haemodynamic responses to noxious stimulus compared to touch stimulus have been reported by Verriotis et al., 2016; and greater spread (FWHM) of the noxious evoked response is related to the larger response amplitude and "smearing" of signal due to low resolution of haemodynamic responses in general i.e. it would be odd if the FWHM was not greater for the larger evoked amplitude.

    1. Nociceptive maps are equally as widespread as touch maps:

    Describing the nociceptive maps as widespread seems inappropriate, unless the touch maps are also described as widespread, which they are not. The authors explicitly demonstrate that both stimulus types (noxious lance and non-noxious touch) evoke activations over areas that are matched in extent i.e. areas of activation do not differ, Fig 3 part b and Figure 3 figure supplement 1 part b. Given that the authors explicitly demonstrate that there is no statistically significant difference in activation extent, both noxious and touch maps should be considered equally widespread. However, this point that nociceptive maps are not uniquely widespread, that this is explicitly demonstrated to be equally the case for non-noxious touch maps, is not made, but is an important point.

    In contrast, the key point is the apparent difference in location of the activated areas i.e. a point about the activity being "mislocalised" rather than being "widespread". The authors show that both stimulus types evoke activations that are centred on the same location (i.e. peak locations do not differ), but the touch stimulus evoked activity extends more medially (toward the location of the foot region), while the noxious stimulus evoked activity extends more laterally (away from the location of the foot region).

    1. The analyses that produce areas of activation in source space (image space) are circular:

    The assessment of changes in [HbO] and [Hb], relative to baseline, in image space is a circular analysis, as it is restricted (informed) by a strongly related analysis in sensor space - "significant peak reconstructed changes in d[HbO] and d[Hb] were identified with a two-tailed t-test (a = 0.01) comparing the peak time point within a 5-second window around the peak latency derived from the channel-wise analysis against the baseline". That is, analysis of the data in sensor space (channel-wise analysis) identified the temporal region-of-interest by comparing evoked activity to pre-stimulus baseline, and this information was used to restrict analysis in source space (image space) to identify regions-of-interest by comparing evoked activity to pre-stimulus baseline. This sequence of analyses is double dipping and therefore circular.

    This issues is outlined in the eLife paper "Science Forum: Ten common statistical mistakes to watch out for when writing or reviewing a manuscript" (, problem number 6 "Circular analysis". This eLife paper also advises pointing to the following paper:

    1. Limitations of the fNRIS technique and of the link between properties of nociceptive maps (widespread/mislocalisation) and poorly directed behaviour are not adequately discussed:

    Tissue damaging painful events elicit cardiovascular responses and infants have underdeveloped cerebral autoregulation. A major potential difference between the touch and noxious stimuli is that the noxious stimuli can elicit blood pressure changes that get reflected in a haemodynamic response measure. This is true for all haemodynamic imaging modalities such as fMRI, fUS, as well as fNIRS. Thus haemodynamic responses are not a perfect reflection of neurodynamic responses, especially in infants. In this manuscript, the authors appear to be aware of this potential issue as in the methods section they use PCA denoising to remove global signals that could be related to blood pressure effects and reference relevant literature: "Optical density changes recorded from all channels (likely related to stimulus dependent systemic physiological changes) were removed using Principal Component Analysis ((Kozberg and Hillman, 2016; Tachtsidis and Scholkmann, 2016); 1 component removed)".

    However, as in all physiological denoising methods, limitations need to be clearly outlined. It is highly unlikely that removing a single PCA component fully eradicates any risk of haemodynamic responses that might not be reflective of neural activity. And as the authors state, the noxious evoked haemodynamic signal appears to be quite widespread including the control channel. While the widespread responses is not statistically significant (see comment 2 above), the fact that the authors are claiming widespread responses extending into control channels seems inconsistent with their claim in the discussion that "The change in the d[HbO] and d[Hb] following sensory stimulation is a measure of neural activity", the NIRS signal "is presumably due to greater depolarisation and spike activity within the activated areas", and their lack of discussion around the limitations of using haemodynamic measures such as fNIRS. In the discussion, fNIRS is contrasted with other modalities such as EEG and fMRI, and fNIRS is claimed to be "ideally suited to a study of this kind". While one can agree that fNIRS has several advantages over techniques such as EEG and MEG, one major issue with fNIRS and (fMRI) is that it is indirect and relies on underdeveloped cerebral vasculature and underdeveloped neurovascular coupling.

    Given that the core claim of this manuscript is that the noxious evoked activity is poorly localised (and maybe widespread), one major risk of using haemodynamic measures for pain studies in infants is that they can easily reflect cardiovascular effects in addition to neural effects. And while removal of a single PCA component was performed, it is not convincing that this simplistic clean-up step removes all risk of non-neural cardiovascular effects.

  3. Reviewer #2 (Public Review):

    Jones and colleagues investigated the topographical similarity of primary somatosensory cortex responses to painful and non-painful touch stimuli in newborn human infants. Their hypothesis was that, as in non-human animal models, responses to non-painful stimuli would be more mature and organized than responses to painful stimuli, which would spread to parts of the somatosensory map other than the affected region. They assessed responses to touch stimuli (repeated, light hammer taps) on the hand and foot as well as responses to a painful heel stick performed for a necessary blood draw. Statistically significant, non-overlapping responses to the hand and foot touches were observed in an organization similar to that seen in adults. The painful heel stick had a similar peak to the non-painful foot stimulation, but the extent of this peak was larger, and the response to the painful stimulus spread into non-overlapping parts of somatosensory cortex, including one channel that had responded to hand stimulation. They conclude that pain responses are more widespread and disorganized relative to innocuous touch sensations early in post-natal development.


    The basic pieces of the methods used here are well chosen to answer this fundamental question about brain development. As the paper points out, fNIRS is an ideal imaging modality here. It provides the necessary spatial specificity and resolution (relative to EEG), while allowing for imaging to occur in flexible environments (relative to fMRI), including the medically necessary, painful procedure used here. The array of optodes was also well designed, providing good density over key regions of interest, as well as a control channel. The authors report changes in both oxygenated and deoxygenate hemoglobin, which helps to support the claim that changes represent canonical hemodynamic responses. I also appreciate the variety of dimensions in which the responses to touch and pain stimuli were compared, including comparisons of the amplitude and spread the peak response using a full width half max calculation, in addition to the number of channels showing significant responses relative to baseline.


    Two elements of the experimental design and implementation make it difficult to tell if the differences in responses to painful and non-painful stimulation are actually a product of different properties of early cortical processing of nociceptive vs. innocuous sensation. First, the touch stimuli were repeated over approximately 10 trials per participant, whereas there was only a single trial of the nociceptive stimulus. In infants, as in adults, repeated stimuli lead to habituation, which is associated with suppression of the neural response. Thus, averaging over 10 trials of the touch stimuli may result in an estimated response that is substantially dampened compared to the response that would be observed to a single, initial trial. This may explain some of the differences in peak amplitude and extent for painful vs. non-painful stimulation.

    Second, when infants moved in the vicinity of a touch trial, that trial was excluded; in contrast, when infants moved after the heel stick, those data were retained. The latter choice was essentially a necessity, as nearly all of the heel stick participants moved. The authors justify the choice by saying that the movements were idiosyncratic and therefore "any associated cortical response would be removed during the averaging process" (line 247). Of course, the responses to such motion would not be removed but rather averaged into the mean group response. It is understandable to argue that this would result in unavoidable noise that could be overcome by strong signal, but under this justification and for parity between conditions, the authors should also retain motion-contaminated trials in the touch conditions. I do also worry that the conditions of the experiment (infants were held closely by their mothers) could result in an underestimate of the motor responses produced by infants in the pain condition, who may have attempted to move their arms or legs more than observed but were held back by their parent, leading to additional pressure and sensation across additional, non-stimulated body parts in that condition. Moreover, the authors do not discuss any differences in gasping or crying between the two conditions, which can also influence cerebral blood oxygenation. (Perhaps this could account for the significant pain response in the control channel, which is largely ignored in interpretation of the results.)

    One final concern regards the channel-wise statistical analysis. The methods state that participants' responses were averaged into a single time course, each time point of which was compared to a baseline distribution (rather than comparing a distribution of responses to a single baseline estimate, or pairing individual participants' response and baseline values). It would seem, however, that the variability in somatosensory responses across participants, and not just variability in baseline, ought to be retained and used to assess statistical significance. This is particularly a concern given the motion confound. For example, strong responses in four infants making large arm or head movements would differentially affect analyses in which 1) those responses only pull up a single average value for each time point, versus 2) those responses also increase the variability of a distribution. Retaining variability in responses to trials would also facilitate directly comparing responses to foot touch and pain in channels claimed to be the locus of the disorganized pain response, which would strengthen the claim that early pain responses are misaligned with the topographical map of responses to touch.


    Given these concerns about the implementation and analyses, it is not clear whether the authors' conclusions are supported by their data. However, adjustments or supplementary analyses could bolster the case for their interpretation that humans, like other animals, initially show exaggerated and disorganized cortical responses to pain in early development.