This study aims to reconstruct the broad climatic -- ice-sheet conditions which enabled the development (and then preservation) of subglacial landscapes along the south-eastern margins of the present-day Greenland Ice Sheet. The study has two primary foci: first, the identification and mapping of 'relict' subglacial valleys using a combination of optical remote sensing and airborne ground penetrating radar, then second, using the configuration of these valleys to explore through ice sheet modelling how they have been formed and preserved - and to explore constraints on the timing of these processes. In doing so, the authors conclude that these features were formed when the ice sheet had a configuration restricted to only a few percent of its contemporary size and was geographically restricted to the mountainous south-eastern area. They conclude that the features were formed around sometime in the range 2.6-7 Ma, so likely well before  the onset of Quaternary glaciation.

My research experience primarily concerns contemporary Greenland surface processes and ice dynamics by fieldwork and remote sensing. I am not active in paleo-glaciology nor in ice-sheet-wide simulations, so I write this review from a more 'generalist' perspective. I want to stress that I welcome rebuttals if the authors feel I've misunderstood their point(s).

Overall I find the manuscript to be in excellent shape. The title is suitable and the paper addresses relevant scientific questions, particularly for this special issue. Substantial contributions on the timing of previous ice sheet extent are reached. The Methods are well described and extensive. I believe that the key areas of potential sensitivity in the modelling have been investigated. I find the results sufficient to support the interpretation. The paper is well-written and overall easy to follow. The figures are of good quality. It was a fascinating read!

I have some specific minor comments as follows:

L119: The authors state that MODIS MoG imagery records the intensity of the reflection of a satellite-emitted radar signal. This is untrue. MODIS is a passive sensor which records visible/near-visible solar radiation reflections.

L125: 'Conversely' is confusing here. Both the preceding sentence and this one focus on shortcomings of the MoG approach, whereas conversely would suggest that we're about to be told something about what it is good for. 'Furthermore' instead?

L136: Please add a further reference that cites the OIB project.

L139-140: Either expand this methods explanation or remove incompletely - there isn't enough information to judge whether the fact that an ML approach was used is important to the present study or not. If it is, expand why, otherwise I suggest removing this methodological detail.

L277: add a reference to Fig. 4g for concerning the thermal state analysis.

L279: (subjective) - I suggest starting a new paragraph here for clarity

L332: is 'mapped mountain valley networks' missing the term 'glacial'? (to distinguish from fluvial)

L480: The first part of this sentence is really the conclusion of the previous paragraph, so perhaps would be better off added there. Then the paragraphs will match points (a) and (b) introduced in L466-469.

L494: would 'also' result in higher rates of mass accumulation and turnover 'there', or similar (i.e. is the intended meaning that the conditions which enable higher rates of mass accumulation in turnover in the EH also cause the ~same conditions in the SH?)

L572-580: I struggled to understand this paragraph. This might be my shortcomings in being able to 'imagine' isostacy, but I think nonetheless that some rephrasing would be beneficial.

In this study, Paxman et al. identify subglacial valleys beneath the highlands of southern and eastern Greenland from MODIS imagery and RES data, and use numerical modelling to constrain the likely age of their incision during mountain-scale glaciation.

The majority of the valleys are presently preserved beneath cold-based/non-erosive ice, indicating their formation is linked to glaciological conditions prior to the growth of a continental-scale ice sheet. Likely timeframes proposed include the late Miocene (7-5 Ma) and/or late Pliocene (3.6-2.6 Ma).

Overall I found the manuscript and figures to be of a very high standard, with the arguments and analyses following in a logical manner. Furthermore, the new data-based constraints provided will serve as useful boundary condition data for future Neogene modelling studies of the GrIS. I have only some minor suggestions/questions below that could help clarify some aspects of the manuscript. 

194: This process for producing a preQuaternary topopgraphy is unclear to me - why are the refilled subglacial valleys included in the isostatic correction if they are subsequently left open? Are any eroded sediments from troughs/marine sectors on the adjacent shelf included in the isostatic response?

344: I think some further clarification on how you calculate these MAD values would be useful e.g., by showing the equation used. I presume its something like (Σ | (Actual) – (Forecast)|)n, but are you keeping the domain size constant when comparing the two binary fields? What is the region being compared? For example, the MAD values in 7u for b&c look identical but the model misfit in terms of extent in reality is very different.

443: Rather state the fraction than force the reader to lookup a figure.

475: Given that ~30% of the glacial valleys are not cold based today and are seemingly being eroded by the present day ice configuration (Figure 5) does this depth contrast between east and south still hold if comparing just the current cold-based valleys? From Fig 6e it seems these coast-facing valleys of the eastern highlands add some skew. 

478: An interesting observation on consistent maximum valley depths - possibly related to negative feedbacks related to sediment evacuation? (cf. Fig 13 Patton et al., 2016). 

517: This asymmetric long-term development - coast v inland - is well-recognised on the Norwegian glaciated passive margin too, and a similar hypothesis was put forward by Kleman et al., 2008 (section 4.1), and Hall et al (2013), during mountain-scale glaciation, and could be useful context here.

544: Arguably I would not have included the PDD-driven models in this Fig 9a given the conclusions of the Plach 2018 paper, which also strengthens your argument here.