Questions Flashcards
(43 cards)
Your entire project rests on the assumption that fetuses can engage in “social cognition.” However, the evidence you cite is extremely limited - behavioral head turns that have been criticized, and a single fMRI study with only 3 subjects. How can you justify a 1.5 million euro project based on such preliminary evidence? Isn’t this putting the cart before the horse?
“While the direct fetal evidence is admittedly limited, the theoretical foundation is stronger than it initially appears. Beyond the studies I cited, converging evidence includes: newborns’ immediate preference for face-like patterns and biological motion, prenatal learning of maternal voice demonstrated across multiple studies, and the prenatal development of cortical architecture for social processing. The innovation of FETAL-MIND is precisely to address this evidence gap using cutting-edge technology. The ERC Starting Grant mechanism is designed for such high-risk, high-reward projects that could fundamentally advance our understanding. If we required extensive preliminary evidence for every study, we’d never make paradigm-shifting discoveries. That said, my phased approach with intermediate milestones ensures responsible use of funds—if Experiment 1 shows negligible effects, we can pivot to focus on the most promising approaches.”
Your normalization approach divides ERF amplitudes by FPSmax to account for variability. However, this assumes that the maximum response during frequency tagging is comparable across subjects and conditions. What if this normalization actually introduces systematic biases that obscure real effects?
“You’ve identified an important methodological concern. The FPSmax normalization could indeed introduce bias if the maximum response varies systematically across conditions. I’ll implement multiple complementary approaches: First, I’ll analyze raw data without normalization to ensure effects aren’t artifacts of preprocessing. Second, I’ll use within-subject z-score normalization as you suggest, which doesn’t assume comparable maxima across subjects. Third, I’ll examine whether FPSmax values correlate with experimental effects—if they do, this would indicate bias. Fourth, I’ll conduct sensitivity analyses using different normalization approaches to ensure results are robust. Most importantly, I’ll preregister all normalization procedures to prevent analytical flexibility. This multi-pronged approach will ensure that findings reflect genuine neural differences rather than methodological artifacts.”
You plan to use multiple analysis approaches (frequency tagging, permutation clustering, neural complexity) across multiple modalities (visual, auditory) and multiple time points (fetal, neonatal). This creates a massive multiple comparisons problem. Your mention of FDR correction seems insufficient - how will you prevent false positives from this analytical flexibility?
I will perform a family-wise FDR correction of multiple tests with each work package. Given that the analysis will be preregistered following WP1, I do not think that the number of tests can or will be excessive. Indeed, it will be considerably less than for a typical neuroimaging study.
Your findings could have profound implications for debates about fetal consciousness and abortion rights. You briefly mention consulting with philosopher Tim Bayne, but this seems inadequate. What concrete steps will you take to prevent misinterpretation or misuse of your findings? Have you considered the possibility that publishing this work could cause real societal harm?
I’m very clear in all my public communications on this topic that fetal consciousness isn’t plausible before the third trimester because thalamocortical synapses haven’t formed yet. The fact that fetal MEG can’t even be recorded before that gestational age of 25 weeks underscores this point. At any rate, I think this type of research is inevitable. If this project isn’t funded, other sites will of course attempt work either with fetal OPM-MEG or fetal fMRI. I’m very comfortable with doing this work not just due to my collaboration with Tim Bayne, but also other philosophers such as Claudia Passos-Ferreira at NYU. Earlier this year, I spoke at a meeting on infant consciousness at NYU hosted by Claudia, and I think the state of this field is very healthy in the sense that scientists are ready to meet this apparent controversy around early consciousness. I’ll develop standardized language for media interactions, vetted by my collaborators, to prevent misinterpretation. I’ll also monitor media coverage and rapidly respond to misinterpretations.
You’re requesting 1.5 million euros as a postdoc with limited experience leading large projects. Your timeline shows three years of data collection, but fetal MEG is notoriously difficult - what happens if recruitment is slower than expected or data quality is poor? Where are your contingency plans?
We recruit pregnant women through the Women’s Clinic at the University of Tuebingen, which is actually one of the top 10 hospitals in the world for OB-GYN medicine. Since 2008, the Tuebingen fMEG Center has recorded 600 fetal MEGs, and the largest study to date recruited 250 pregnant women with 60 recordings per year. We have a strong track record for recruitment and data collection. [Add that Helmholz pays for helium]
As far as my leadership credentials, my first senior author publication is currently under review at PLOS Complex Systems, and I currently supervise two students and a more junior postdoc in the lab.
Your proposal focuses exclusively on term fetuses, yet preterm infants born at similar gestational ages are readily accessible for study with proven neuroimaging techniques. Why not validate your paradigms first in preterm infants where you could use high-density EEG or even fMRI? This would provide crucial proof-of-concept data with better spatial resolution and established methodologies.
Sure, well the main reason is that preterm infants aren’t the best models of term-equivalent fetuses. Preterm infants generally leave the womb due to some threat to their health in the uterine environment, and the process of birth plus the ex-utero environment both greatly change the brain in terms of arousal, behavior, and perhaps perceptual capabilities. I don’t think that positive results from preterm infants would necessarily tell us much about term-equivalent fetuses or what’s actually happening in the womb. Regarding fMRI, fetal neuroimaging with fMRI is already possible and perhaps even easier than it is for preterm infants. In fact, the womb is a much more controlled environment than the cradle. Fetuses don’t cry in the womb, which actually can make studying them easier in some ways.
Your project assumes that detecting neural responses to social stimuli indicates “cognition” rather than mere sensory processing. What specific features of the neural responses would distinguish cognitive processing from automatic sensory responses? How will you differentiate between subcortical reflexive responses and cortical cognitive processing given your technical limitations?
The control stimuli will distinguish social processing from non-specific sensory processing. If the fetal brain’s response isn’t specific to the social stimuli but is also evoked by other sounds or visual patterns, then it would indeed appear to be mere sensory processing. So, my hypothesis around prenatal social cognition is certainly a falsifiable hypothesis. Subcortical activity doesn’t get recorded by MEG, which is only sensitive to cortical responses. If I see an MEG response, I can be confident that isn’t not a midbrain reflex but actual cortical activity.
Your sample size calculation depends on effect sizes from Experiment 1, but what if these initial effects are inflated due to publication bias or sampling variability? Would it not be more prudent to power your study based on conservative effect size estimates from the broader infant literature?
“You’re right to highlight this issue. Pilot studies are notorious for producing unstable effect size estimates due to sampling variability. I’ll address this by committing to recruit at least 120 participants for WP2 regardless of WP1 results, which provides 80% power to detect even small-to-medium effects (d=0.35). If Experiment 1 yields large effect sizes, I’ll treat these skeptically as potential overestimates. Conversely, small effect sizes in Experiment 1 would indeed be informative, suggesting we should focus resources on the more promising paradigm. I’ll also preregister decision rules for paradigm selection based on effect size confidence intervals rather than point estimates, ensuring transparent and statistically sound progression from pilot to confirmatory phases.”
How does the fetus represent the stimulus? [The dreaded NYU question]
“The fetus represents these stimuli through neural encoding of specific features - for visual stimuli, the top-heavy spatial configuration, and for auditory stimuli, the intonation patterns that distinguish human voices from other sounds. This representation operates at a systems level that doesn’t necessarily require conscious awareness, reflecting emerging categorical distinctions in how the brain processes these social versus non-social acoustic and visual features. I should note that questions of neural representation remain theoretically complex, and our approach focuses on measurable differential responses rather than making strong claims about the nature of fetal mental representation.”
The control group of posterior-facing fetuses introduces a confound - these fetuses differ not only in visual stimulus exposure but also in position, which could systematically affect MEG recordings. How do you control for this?
The MAIN control for the face-like visual stimulus is in the form of inverted and random control patterns. FETAL-MIND doesn’t rely upon the backward facing fetuses for a control, but if the backward fetuses don’t show a different response than the forward facing fetuses, then that would definitely be reason for skepticism.
Your power analysis is circular - you plan to determine sample size based on effect sizes from WP1, but this approach cannot guard against underpowered initial results leading to false negatives. What minimum detectable effect size would you consider meaningful?
I want to see moderate effect sizes, Cohen’s d >= 0.3 … anything less than that, and I need to consider dropping that stimulus paradigm from WP2.
[Note, d = 0.3 gives 80% power with n = 90 for paired samples t-test]
The cortical plate in 32-week fetuses is still undergoing significant reorganization, with the subplate being the dominant structure. Since MEG primarily detects currents from pyramidal neurons in layers 2/3 and 5, how can you be certain your signals originate from the cortical plate rather than the transient subplate neurons? Doesn’t this fundamentally challenge your interpretation of “cortical” responses?
Two points:
1) It’s unlikely that MEG would detect anything beneath the cortical plate. It might be theoretically possible, but it’s not very plausible given that MEG tends to detect superficial cortical sources, more so than other techniques like EEG, as magnetic fields follow an inverse cube law that makes detection of deep brain signals very challenging, even more so for fetal MEG where there is already some distance between the fetal head and the MEG sensors.
2) While it’s true that MEG preferentially detects superficial sources due to the inverse cube law, the distinction between cortical plate and subplate activity at 32 weeks is actually less problematic than it might appear. By 32 weeks, the cortical plate has already established its six-layer structure, and thalamocortical afferents have reached layer 4. The subplate, while still present, is beginning its regression. More importantly, the evoked responses we’re measuring require thalamocortical input, which by definition involves cortical plate neurons. The synchronized postsynaptic potentials generating our MEG signals must originate from the radially oriented pyramidal cells in layers 2/3 and 5 of the cortical plate, as subplate neurons lack this columnar organization.
Cortical folding accelerates dramatically between 32-37 weeks, potentially changing the orientation of pyramidal neurons relative to MEG sensors. A fetus recorded at 32 weeks might have smooth cortical regions that become sulcated by 37 weeks. How will you account for these dramatic anatomical changes when comparing across gestational ages?
You’re right that cortical folding introduces variability, primarily because we’re recording across a wide gestational age range. A 32-week fetus has relatively smooth cortex while a 37-week fetus shows extensive secondary sulci. However, this is addressed by including gestational age as a covariate in all analyses. The bigger source of variability is actually fetal head position - the same cortical region might be oriented differently relative to sensors across recordings. This is why we spatially average across the 10 channels with strongest signal rather than attempting precise localization.
MEG detects magnetic fields from synchronous postsynaptic currents in pyramidal neurons. However, fetal neurons show much higher spontaneous firing rates and less precise synchronization than mature neurons. What’s your minimum threshold for cellular synchronization needed to generate detectable MEG signals, and how did you calculate this given the immature state of fetal neurons?
“The higher spontaneous firing in fetal neurons actually helps by increasing the pool of potentially recruitable neurons. The key is that sensory stimulation provides external synchronization that overcomes the intrinsic variability. Prior work has shown clear evoked responses, so whatever the precise threshold, fetal neurons achieve sufficient synchronization for detection. Future computational modeling could refine these estimates, but isn’t necessary for interpreting our experimental results.”
If your hypothesis about innate social templates is correct, you’re essentially claiming that evolution has created a “critical period” that opens before birth. What molecular mechanisms would support such early critical period opening? Most known critical periods (like ocular dominance) require sensory experience to trigger. How do you reconcile this with the absence of face exposure in utero?
“You’re right that I’m not proposing a classic critical period. Instead, I’m suggesting experience-expectant mechanisms where evolution has prepared circuits to expect certain inputs. Unlike ocular dominance, which requires visual input to segregate, face and voice templates could emerge through intrinsic developmental programs. The innovation is that these templates become functionally testable in the third trimester, not that they require fetal experience to develop.
The molecular machinery would involve morphogen gradients (like ephrins and netrins) that guide connectivity between fusiform areas and earlier visual regions, combined with activity-independent mechanisms like molecular recognition between pre- and post-synaptic partners. Genes like FOXP2 for vocal processing and face-selective patches guided by retinotopic gradients could establish basic templates.”
The budget seems to assume the Tübingen fMEG center provides unlimited free access to their unique facility. What’s your backup plan if they start charging for scan time or if their funding situation changes during your 5-year project?
You also mention multiple software needs (MATLAB licenses, statistical software) and data storage (800TB mentioned in the proposal). Where are these costs? A single MATLAB license with necessary toolboxes can cost 5,000+ EUR annually.
This won’t happen, as the main expense, helium, is paid for by the Helmholz Center Munich, and the hard drives are already paid for and maintained through other funds. Similarly, MATLAB licenses are already paid through other funds.
You don’t show any examples of actual fetal MEG traces or the cardiac artifact removal process, despite this being central to your methodology. How can we assess the quality of your preprocessing without seeing raw versus cleaned data?
Given page limits, I prioritized experimental design over preprocessing details, but I should have included at least one figure showing the dramatic transformation from raw to cleaned data.
Let me briefly describe what you would see: The raw fetal MEG signal is completely dominated by the maternal cardiac artifact. The fetal cardiac signal is visible but 10x smaller. After our FLORA/FAUNA pipeline, both cardiac signals are cleanly removed, revealing the underlying fetal brain activity at the fT scale.
You want to record an N290 analog with face-like stimuli in the fetus, but the N290 is a mainly posterior ERP probably generated in the fusiform gyrus. Because the stimulation device is between the maternal abdomen and the MEG sensors, and because only the group of forward facing fetuses can see the face-like patterns projected into the womb, you will only record frontal cortical signals in these fetuses. So, an N290 analog is completely infeasible. How do you address this?
“The forward-facing orientation does not necesaarily limit us to anterior cortical signals, as the fetal brain is small enough that it can often be modeled as a point source for all practical purposes. If a posterior response is strong enoigh, it might still be measured in a forward facing position.
That said, the infant Nc or “negative central” component, which peaks around 400-800ms and is generated in medial-anterior regions, is perhaps more relevant to our paradigm. The Nc reflects attentional allocation to salient stimuli and, crucially, shows differential amplitudes for social versus non-social stimuli even in very young infants.
The anatomical alignment is fortuitous—the Nc’s anterior generators are exactly what we can access in forward-facing fetuses. This component represents domain-general attention rather than specialized face processing, which aligns with our theoretical framework. We’re not claiming fetuses have mature face recognition systems, but rather that face-like configurations might capture attention differently than control stimuli.
Richards and colleagues have shown that the infant Nc amplitude correlates with heart rate deceleration during sustained attention, which connects directly to our WP3 hypothesis about convergent physiological markers. If we observe both anterior slow-wave responses and cardiac deceleration to social stimuli, this would suggest genuine attentional engagement rather than simple sensory discrimination.”
It may be the case that infants better attend to female voices. Will you plan to use only female voices, or both male and female voices?
I don’t think there’s strictly right or wrong answer here. Because there are benefits of both approaches, the benefit of using female voices is that they’re attended to better by infants. But the benefit of using both voices, male and female is that you would have better generalizability of results, that being said i’m inclined toward only using female voices at the moment.
Why do you think the fetal heart rate will decelerate in response to social stimuli? Has this been shown before in fetuses?
This is based on prior studies in infants. For instance, Petrie Thomas (“Pet-ree Tah-mah”) et al. showed infant heart rate deceleration in response to novel toys in both full-term and preterm infants. (If pressed further) I also think heart rate deceleration is important to show that this is a response to conspecific cues and not, say, a threat detection response.
The fetus’s responses to voices might actually be indirect responses to changes in the maternal heart rate triggered by voices. How will you handle that?
We can simply provide the mother with music to listen to with MEG compatible earbuds. These use air filled tubes that conduct air inside of an earplug-like material. That way, the mother cannot hear the stimuli well and she is distracted. We can also of course check whether the mother’s heart is actually changing in response to the voice stimuli. If the mother’s heart rate is responding, we can try better masking, e.g., using white noise or brown noise that the mother listens to with the earbuds.
Will you monitor the fetal behavioral/sleep state? Why or why not?
Fetal behavioral state can be deduced offline using heart rate variability. Lorenzo Semeia, a postdoc here at the fMEG Center, showed that HRV is more useful than fetal movement for classifying fetal behavioral states. Fetal MEG already records fetal heart activity, and so we can easily covary for HRV as necessary in the multiple regression models. In previous work by Moser et al., the global prediction error effect in fetuses was dependent upon HRV, so some fetal neural effects certainly seems to be contingent upon HRV and behavioral state. I don’t like to think of these states as sleep because it’s a complicated question if and how often the fetus is sleeping. Indeed, it’s not really necessary to treat these as discrete states. HRV is a continuous variable and would be entered into models as a continuous covariate.
Since the fetus is likely asleep, how do you expect to see exactly the same response to social stimuli in fetuses and newborns?
I don’t expect to see exactly the same response, I just expect to see a correlation in response amplitudes even if the nature of those responses is slightly different. The variance in fetal arousal state, as indexed by HRV, is sufficient that past fetal MEG studies have been successful in recording neural responses, including at least one, the global prediction error, which is thought to require conscious awareness.
Will you use both female and male voices in the auditory stimuli?
“This is a challenging methodological decision with trade-offs either way. Including both male and female voices would add substantial complexity to our design and analysis, requiring larger sample sizes to detect effects across voice categories.
Given that infant research shows stronger attention to female voices - likely due to prenatal familiarity with maternal speech patterns - we’re planning to use female voices exclusively for our main experiments. This controls for the substantial acoustic differences between male and female voices that could confound our social versus non-social comparisons.
While this limits generalizability across voice types, using only female voices allows us to focus our statistical power on the core question of vocal versus non-vocal processing, rather than diluting effects across gender categories. As a legacy of the project, later work might test generalization to male voices in follow-up studies once we’ve established the basic phenomenon.”
