Article activity feed

  1. Evaluation Summary:

    This paper explores the use of 2D high-resolution template-matching (2DTM) to locate and discriminate highly similar macromolecules within cryo-EM images of focused ion beam-milled cells. It demonstrates that differences in the 2DTM signal-to-noise ratios for located targets against multiple search templates can effectively segregate a mixed population of similar structures, as well as present a formal analysis strategy for probabilistic assignment of species within the mixed population. Because the identification of distinct structural states of macromolecular complexes inside the cell is a fundamental problem in 3D visual proteomics, this paper will be of broad interest to both structural and cell biologists.

    (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. Reviewer #1 and Reviewer #3 agreed to share their name with the authors.)

    Was this evaluation helpful?
  2. Reviewer #1 (Public Review):

    This paper provides a progress report on methods development that was initiated previously by the same authors to identify macromolecular complexes in cryo-EM images of cells. Whereas others have proposed to perform this task in 3D reconstructions from tomographic tilt series, the method discussed here (2DTM) uses template matching with known reference structures against individual 2D projection images. This concept was introduced in previous work. In the current paper, the authors show that 2DTM can be used to classify distinct molecular populations. They demonstrate this by distinguishing cytoplasmic mature 60S ribosome particles from a nuclear pre-60S population. They also define a maximum likelihood metric that assigns a probability for each particle belonging to each class. The paper reads a bit dense, and one could discuss how big the advance is over the previous work by the same authors. But the general topic (of being able to identify distinct particle populations in cells) is an important one.

    However, I think one major concern needs to be addressed much more openly in a revised version of the paper: that of potential model bias of incorrect references. The single-particle field previously suffered the problems of Einstein from noise can cause with the debacle around an incorrect HIV trimer reconstruction that was the result of picking particles with a high-resolution reference. I realise that the problem here is a different one, but similar problems of model bias may exist. In fact, the observation on page 18 that the reconstruction from the picked particles was estimated by FSC to be 3.5A, yet the resulting map had to be filtered at 10A to limit the noise, is a strong indication that model bias does play a large role in the identification of particles. This bias must affect the measured SNR scores, and thus the metrics presented. It also suggests that part of the identified picks may in fact not be true 60S ribosomes, but false positives. This would then affect the conclusions drawn. If the authors disagree (and I suspect they do), they should set out clear arguments for their case. Also, they should discuss how potential overfitting or model bias would affect their new metrics for particle classification in the discussion. Currently, the only reference to the dangers of overfitting is on page 16, merely referring to their previous paper.

    Perhaps points for additional discussion could include:

    1. In the light of overfitting, I was wondering whether one could detect the 80S ribosomes also through 2DTM using the 40S subunit as a reference.
    2. The authors have been wise in selecting the 60S ribosome as a test case. Probably, because of its size and RNA content, for many instances of this complex the SNR is sufficiently high for detection. However, if less careful authors would choose a smaller target, what would happen? What would be the pitfalls and how could they be avoided?
    Was this evaluation helpful?
  3. Reviewer #2 (Public Review):

    Lucas et al extend their previous studies on the usage of 2DTM in cells (cfr Lucas et al, elife 2021). In this study, they detect ribosomes in FIB milled lamellae using a set of templates representing different maturation stages.
    In agreement with previously published data, they find that templates representing a more or less mature ribosome preferentially detect particles in the cytoplasm and in the nucleus respectively. This nicely shows that 2DTM has the sensitivity to distinguish between similar molecules. In addition, they use the prior knowledge that the different maturation intermediate are spatially separated to develop a maximum likelihood approach to assign the probability that each particle belongs to one or the other class.

    The main caveats of this study are that:
    1. It is unclear how broadly the method can be applied, for particles smaller or with more subtle variations than ribosomes, and most importantly in the case of particles which are not spatially separated and for which relative ratio are completely unknown.
    2. The authors had already shown that 2DTM can be applied in situ in previously published work. If classification is limited only to cases where the ratio between particles numbers in the different states is known, then the real progress with respect to previous work is somewhat marginal.

    Was this evaluation helpful?
  4. Reviewer #3 (Public Review):

    Lucas et al. expand upon their prior work using 2D high-resolution template-matching (2DTM) to localize macromolecules directly in cells. This clearly presented work contains multiple key highlights using the Saccharomyces cerevisiae 60S maturation process as an example. It demonstrates that focused ion beam (FIB)-milling preserves sufficient high-resolution (better than 4 Å) information for the 2DTM to effectively locate macromolecules in the dense cellular environment. In addition, it demonstrates that the classification of the detected macromolecules can be effectively determined by comparison of the signal-to-noise ratios obtained with 2DTM against templates with relatively minor differences. Furthermore, the authors detail a maximum likelihood approach to specify the confidence of the class assignment for a macromolecule within a mixed population. The authors take advantage of extensive prior knowledge of the 60S biogenesis process to thoroughly evaluate and demonstrate the utility of the 2DTM methodology and accompanying classification strategy.

    2DTM has great potential to motivate a broader adoption of cryo-EM for those more interested in robust localization of macromolecules of known structure rather than de novo high-resolution structure determination through conventional averaging approaches. Conventional averaging approaches for cryo-EM data notably suffer at the level of classification for which the results can vary greatly based on choice of a multitude of parameters. The classification strategy presented here for 2DTM should be reproducible and the parameter choice (i.e., priors) more straightforward.

    Was this evaluation helpful?