This is an important study addressing the mechanisms of variant gene expression and switching in the malaria parasite Plasmodium falciparum. The work provides solid evidence supporting the existence of a non-random, highly structured switch pathway for var genes and identifies one var gene, called var2csa, as a sink node in this network. The findings consolidate previous observations and further the understanding of the enigmatic mechanism that underlies the regulation of the var gene family and antigenic variation in P. falciparum, which is paramount for immune evasion and acquired immunity and further influences malaria pathology.

[Editors' note: this paper was reviewed by Review Commons.]

Thank you for submitting your article "A coordinated transcriptional switching network mediates antigenic variation of human malaria parasites" for consideration by eLife. Your article has been reviewed by two reviewers (including one from the initial evaluation at Review Commons), and the evaluation at eLife has been overseen by a Reviewing Editor and Dominique Soldati-Favre as the Senior Editor.

Based on the previous reviews and the revisions, the manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below.

Recommendations for authors

1. For many clonal lineages, only pie charts are provided for visual inspection and categorization into 'single' or 'many'. However, this also appears to be largely dependent on overall var gene expression, as in Figure S1A, subclone NF54-5 has dominant expression of a single variant and low overall var gene expression and was therefore classified as 'many'. The authors should provide expression levels as displayed in Figure 1C for all subclones analyzed in the study. These graphs could be displayed next to the corresponding pie chart, or as supplemental figures.

Importantly, an unbiased approach to categorizing parasites into different states would be desirable. Perhaps the combination of a diversity index and total var gene expression could provide sufficient discrimination.

2. The transcriptomics data reveal that "manys" progress through the first half of the replicative cycle faster, yet the "singles" catch up in the second half, and thus both populations have equivalent replication cycle lengths. The authors should discuss further why 'many' parasites have no growth advantage over several parasite generations, despite faster ring stage progression. Do the 'single' parasites catch up with this lag later? Giemsa smears, in addition to growth curves, would be useful to better document progression through the IDC of 'single' versus 'many' parasites.

3. The authors should discuss how the switch model proposed by Recker et al. could work with only a single dominant sink node that is barely inactivated during the cultivation of the parasites in vitro. In fact, what about the PF3D7_0421100 gene, which is also frequently activated and stably expressed in many subclones in different generations (Figure 2B)? Could this also be a sink node?

4. Reviewers appreciated that the authors have made an effort to improve the readability and presentation of the data. However, they also noted that there is still room for improvement.

a) The labeling of the heatmaps and bar graphs is not consistent with respect to the order of var genes and the var groups are labeled twice in the heatmaps (largely on the left side and after each gene).

b) There are some inconsistencies: cultivation days for the same clonal lines are not identical in Figures 2A and 3A; clonal line V2dis2 is classified as "many" in Figure 4A but as "single" in Figures 4E and F.

c) Why do the authors not show var2csa expression in the wild-type heat maps, but in the pie charts?

d) The phylogenetic trees of clones and subclones are partly redundant, but an overview including all clonal lineages, e.g. also clones 1, 2, 4, and 5, is still missing (could be included as a Supplemental figure).

e) Var1 should not be labeled as group D, which has been shown not to exist. The authors also do not explicitly designate var3 type genes, and the reviewers suggest including var1 and var3 in group A and more accurately labeling them var1 and var3 variants.

f) The similarity of the minor transcripts between Figure 2A and C is difficult to judge from the heat maps. Why did the authors not include the subclones from Figure 2C in the Bray-Curtis dissimilarity analysis shown in Figures 4C and D?

5. After disruption of the var2csa promoter, do the authors still see var2csa expression, or can they confirm its absence?

We now appreciate that the original naming scheme applied to the clones was confusing and difficult to follow. We have therefore renamed all the clones described in the manuscript and used names that reflect their origins. We hope this clarifies the logic and progression of the paper.

We appreciate the concerns of the reviewer and have modified the figures appropriately. As suggested by the reviewer, the heatmaps in Figures 2 and 4 are no longer arranged using hierarchical clustering and instead have been restructured according to ups type. This provides a standardized structure and enables easy comparison of all the figures, as the reviewer had hoped. Patterns are now discernable. We thank the reviewer for this suggestion. The reviewer was also concerned about the calculation of column z-scores derived from relative copy numbers in the construction of the heatmaps, particularly if the amplicons from each var gene are not of equal length. However, the primer set designed by Salanti et al. (Mol Micro, 2003) was specifically designed and tested to enable absolute quantification. Because we intend to compare patterns across the different clones, we prefer this method for creating the heatmaps. We now specifically mention this in the Results section and the Methods section (see lines 131-134 and 601-608).

The clones chosen for the var profiling were not arbitrary, although with the previous naming scheme this might have been difficult to follow. We hope the new naming organization will make our logic clear. The timepoints for analysis are constant in Figure 2A. The reviewer is correct that timepoints shown in Figures 3 and 4 are not constant. While perhaps not ideal, this reflects the evolution of the experiment over the years in which the project was conducted and does not alter the conclusions of the experiments. The time points are now clearly marked in each figure.

The var2csa deletion (via chromosome truncation) resulted from multiple independent events that were previously described in detail in Zhang et al. PLOS Bio, 2019. Clones derived from three of these independent events were fully characterized in the previous paper and were used in the current manuscript. All displayed the same phenotype. The CRISPR-mediated integration was repeated in both 3D7 and NF54, and all clones provided the same phenotype. We have added to the text to make this clearer (see lines 260-271).

The reviewer makes a good point here that we did not discuss in the original version of the manuscript. In previously published work, we had deleted several var genes in alternative positions in the genome and did not observe the switching impairment described here. In addition, the 3D7 reference line that is frequently used throughout the world has subtelomeric deletions of var genes on both ends of chromosome 14 (Calhoun et al., 2017), but appears to undergo var transcriptional switching. We now discuss this and provide citations in the Discussion section (see lines 478-482). We also added two additional control lines to the RNAseq analysis (Figure 5) in which we disrupted a rif gene near var2csa. These lines clearly display a wildtype phenotype as expected. We also now note in the discussion that reference 31 (Mok et al) concluded that switching to var2csa was independent of genomic rearrangement (lines 461-463) and mention the work of Duffy et al. (lines 458-461), which similarly reported a genomic rearrangement adjacent to var2csa that resulted in var2csa activation leading them to conclude that the effect resulted from the cis-sequence surrounding var2csa, a hypothesis consistent with our model indicating that var2csa is highly competitive. We think these observations are entirely consistent with one another and together lead to a more refined model of how the var network is coregulated.

The reviewer describes a very plausible model suggesting that energy conservation from not expressing PfEMP1 could lead to convergence to var2csa. The more rapid progression through the first half of the replicative cycle by “manys” is consistent with this hypothesis. To address this possibility, we generated growth assays for “single” and “manys” as well as parasites in which var2csa had been deleted and detected no difference in growth rates (see new Supplementary Figure 1). We also observed that while “manys” progress through the first half of the replicative cycle faster, the “singles” catch up in the second half, and thus both populations have equivalent replication cycle lengths. We have added a flow cytometry analysis of DNA replication to Figure 5 to show that all clones replicate their DNA in a similar time frame suggesting that the more rapid progression through the early stages of the asexual cycle does not slow replication. We now discuss these ideas in the Discussion section (see lines 482-488).

The reviewer’s suggestion regarding creating a point mutation in the start codon of the var2csa uORF is an excellent idea and something we had previously done for a different line of investigation. This generates a profound phenotype unrelated to what we are describing in this paper and thus would be uninformative for the current manuscript. This will be described in an independent submission.

This has been corrected.

qRT-PCR reactions were normalized to seryl-tRNA synthetase (PF3D7_0717700). This is now stated in the Methods section (see line 610).

As described above, all the heatmaps have been restructured, and the gene annotation numbers are now correct and organized according to ups type.

We apologize for the confusion. Figure 3B and C display two alternative methods that both disrupted the function of var2csa as a regulator of var gene switching. Indeed, the transcriptomes of parasites derived using both of these methods are analyzed in Figures 4 and 5. We have improved the description in the text to make this clear (see lines 260-271).

We agree that the original notation in the figure was unclear and confusing. As mentioned above, we have restructured the naming scheme for all the clones described in the manuscript. In addition, Figure 4A now separates the analysis for parasites derived from the different methods used to disrupt var2csa.

We now arrange the gene expression profiles according to ups type rather than hierarchical clustering, therefore this section has been rewritten. We now emphasize that upon disruption of var2csa, the pattern of minor transcripts is more divergent and dispersed over time or when comparing closely related subclones, as quantified in Figure 4C.

The reviewer is correct and we have corrected the figure as suggested.

We agree with the reviewer that it would be best to have the two figures using comparable time points, however the figures were derived from different previously published datasets, and thus we are unable to convert the data. We have now separated the analysis into two separate panels to avoid confusion.

We agree with the reviewer and have corrected the figure as suggested.

At the suggestion of one of the other reviewers, the analysis of the previously published single cell RNAseq data has been removed.

We agree and have added appropriate references to this paragraph (see line 75 and references 19-21).

We have modified this description as suggested by the reviewer (see lines 75-78).

We agree and have added appropriate references to this paragraph (see lines 78-80 and references 25 and 26).

We agree with the reviewer that any discussion of expression patterns in vivo must consider immune selection. We have now added a paragraph to discuss this topic to the Discussion section (see lines 505-517).

As described in our responses to Reviewer 1, we have modified the figures that display var activation patterns to include ups-type (see annotation of ups type in the clones shown in Figure 1D, for example), and all the heatmaps now have a uniform arrangement of genes arranged by ups-type. This enables the reader to consider the characteristics mentioned by the reviewer.

We considered this, and recreated the graphs with a single scale, however the white space became quite large and we prefer the reduced scale to give the reader the most resolution of the data.

We agree with the reviewer that the differences between wildtype singles and manys should be similar to the differences we observe between the var2csa mutant lines, albeit smaller. At the request of the reviewer, we now separate the wildtype “singles” and “manys” in Figure 5F. As predicted, this analysis shows the wildtype lines do in fact display intermediate levels of gene expression, consistent with a shift in cell cycle progression. We thank the reviewer for this excellent suggestion and agree that this is likely to make it easier for readers to understand this concept.

In this sentence, we were specifically referring to on-rates, rather than overall switch rates. We have modified the sentence to make this clear (see line 326).

We have added an additional dataset to Figure 1 which incorporates data from a large number of wildtype subclones. Figure 1D has been greatly expanded to include 41 closely related, clonal lines, and data included in a new Figure 1E includes data from a total of 74 lines.

Figure 1 was intentionally kept simple to illustrate that the single and many states are reversible. However, given the reviewer’s comments, we have now expanded Figure 1D to show a larger set of wildtype subclones through multiple “generations”. We have also added quantification of the number of subclones displaying each phenotype and the level of var gene expression in the lines displaying either the “single” or “many” expression profile (Figure 1E).

We have added analysis of additional recently subcloned populations to figure 1D and E, including statistical analysis. We also analyzed several clones by RNA-seq in Figure 4F, and the trend is quite clear. Regarding how long these states last, this is shown in Figure 3A for wildtype parasites and in Figure 4A for var2csa deletion lines.

Most of the experiments described in the manuscript were performed in 3D7 due to the availability of tools to easily, rapidly and quantitatively examine the full var repertoire. In the absence of these tools, it is difficult to use the same approach in alternative isolates. However, Recker et al. also observed the single and many states in IT, a different genetic background, and found the pattern to be comparable to what they observed in NF54 and 3D7 (ref 22). This is where the concept of “single” and “many” comes from. We now explain more clearly that we are applying an optimized qRT-PCR method in 3D7 to investigate an observation originally made in IT (see lines 127-133). However, to ensure that our observations are not a result of exceptionally long-term culture of 3D7, we obtained a stock of NF54 (the original isolate from which 3D7 was derived) that has been cultured for many fewer generations and that maintains many wildtype qualities (knobs, sexual differentiation, etc). Cloning of this culture also yielded “single” and “many” lines with similar characteristics and frequency as our original experiments using 3D7. We now include these data in Supplementary Figure 1. We also now mention that two other labs have similarly observed dramatic changes in var expression levels and heterogeneity in both NF54 and IT (references 29 and 30), indicating that this phenomenon is not specific to our 3D7 line (see lines 151-157).

The reviewer mentions many clever experiments and hypotheses for further deciphering the mechanism by which var2csa serves as a node in the switching network. Many ideas mentioned here we have also considered and are designing experiments to test, although incorporating all of them into the current manuscript is not possible given time and space constraints. The concept of promoter competition, with var2csa being a strong promoter, is now described in the 2^nd paragraph of the Discussion section, and we added a discussion of the Duffy et papers from 2009 and 2016 (see lines 458-461). We did not enrich our populations for activation of specific var genes, however we noticed no differences when we examined populations expressing different dominant var genes.

We have greatly expanded the methods section to provide details into how we synchronized our parasite cultures to ensure that all were entering the replicative cycle simultaneously (see lines 589-598). We performed the experiments in the replication cycle immediately after synchronization, thus the differences in gene expression reflect how rapidly the ring stages progress, rather than the populations becoming out of sync with one another over time. We also find that the overall cycle progression is similar in all lines, which is also reflected by the timing of DNA replication (Figure 5G) and in the overall growth rates of the parasites (Supplementary Figure 2). These data suggest that the more rapid progression through the early stages of the replicative cycle does not slow parasite replication. The reviewer brings up the question of whether reduced var/PfEMP1 expression might result in faster progression through the cycle (as also mentioned by reviewer 1). While this is an interesting and plausible hypothesis, we don’t detect the expected change in replication rates when comparing “singles” vs “manys” (Supplementary Figure 2). We now describe this in the Discussion section (see lines 482-488).

The reviewer is correct that we could use an additional method to estimate progression of the replicative cycle in each of the lines, although we find this unnecessary. The point we are trying to make in this analysis is that the “manys” progress more rapidly through the ring stage than the “singles” and that this difference explains the differences in the transcriptomes. Exactly at what point in the cycle the samples were obtained is less important. We feel the analysis we have included supports this conclusion well. Importantly, all 11 clonal populations fit this model precisely, providing additional confidence in this conclusion. We have now added growth curves (Supplementary Figure 2) as well as analysis of DNA replication by flow cytometry (Figure 5G) to further define the differences in cell cycle progression.

We agree with the reviewer that the data from the Howick paper are not ideal for the analysis we would like to perform. These limitations cannot be overcome with the current dataset, so we have taken the reviewer’s advice and omitted this analysis.

As mentioned above, we have removed this section.

We have reexamined the text with respect to the reviewer’s suggestions and made many changes. However, with respect to mentions of model systems like the olfactory receptor gene example, we prefer to keep the citations as is, because we think it helps to bring clarity to a very complex concept. Given that many readers are likely to not be experts in the concepts of mutually exclusive expression and transcriptional gene choice, we feel that giving ample examples and repeating them when necessary is helpful.

These are excellent questions. We would prefer to defer to mathematical modelers to determine how current models could be modified to accommodate a single, dominant sink node. This is not our expertise. How var2csa is inactivated is an active line of research ongoing in our lab. We suspect that the unique structure of the gene, including the uORF and its encoded protein, are likely involved, but this will require extensive additional work beyond the current manuscript to definitively determine.

These have been performed and added to the manuscript (see Supplementary Figure 2) as well as flow cytometry measuring DNA replication (Figure 5G).

The original identification of the single-many-single (SMS) model was from qRT-PCR and Northern blot data using the genetic isolates IT and 3D7 (Recker, 2011). In our paper, we attempted to verify and further examine this model using a somewhat refined quantitative qRT-PCR method. This employs a specific var primer set originally designed by Salanti et al., 2003 that has been modified and improved as the genome sequence of 3D7 has been updated. This is a unique primer set because it was developed to allow absolute quantification (see similar comments to reviewer 1). Unfortunately, this primer set is only suitable for analysis of 3D7 or NF54. We now explain this more clearly in the manuscript, noting that Recker et al. observed similar “many” and “single” expression patterns in the IT parasite line (see lines 127-135). We also note that Merrick et al. (2015) similarly observed stark differences in var expression levels using independently derived clones of NF54 and 3D7 (see lines 149-152), as did Janes et al. (2011) working with the IT line, indicating that this phenomenon is not unique to the lines we describe. Nonetheless, to ensure that our observations are not a result of exceptionally long-term culture of 3D7, we obtained a stock of NF54 (the original isolate from which 3D7 was derived) that has been cultured for many fewer generations and that maintains many wildtype qualities (knobs, sexual differentiation, etc). Cloning of this culture also yielded “single” and “many” lines with similar characteristics and frequency to our original observations using 3D7. We now include these data in Supplementary Figure 1.

- A limitation of the manuscript in its current form is the exclusive focus on the mRNA transcriptional level and the lack of protein level data. Do the observed differences at the var gene transcriptional level correlate with less PfEMP1 protein at the parasite bulk level? This would allow to determine the effects of "single" and "many" status on PfEMP1 levels and whether var2csa needs to be translated to act as a sink node, thus also providing insight into the mechanism, which would expand the scope of the study and potentially increase its relevance. If this is beyond the scope of this study, the authors should discuss these points in their revised manuscript.

We agree that expanding the study to PfEMP1 expression levels would be an excellent next step, and we intend to pursue these studies in the future. However, the current manuscript is focused on the var transcriptional network, coordination of var gene activation and the identification of var2csa as a potential sink node. In our view, these are important findings independent of protein expression levels. Nonetheless, we have added discussion of PfEMP1 expression to the revised Discussion section (see lines 482-488 and 505-517).

The reviewer is correct that panning has the potential to determine if it is possible for var2csa-disrupted parasites to undergo switching, however we already have data indicating that this can occur at a low frequency. For example, in Figure 4 A and B, parasites expressing different var gene profiles can be observed, however as suggested by the reviewer, this appears to occur at a much lower frequency. We have modified the text to make this clear (see lines 295-298).

As mentioned above, the methodology we employed in this study is heavily reliant on the primer set of Salanti et al., which is only suitable for NF54 and 3D7. However, we now describe and cite published work from others that used both the NF54 and IT lines, and which displayed a similar phenomenon (lines 127-135, 151-154 and references 22, 29 and 30), and we have added a similar analysis from the original NF54 isolate that has not been in culture for as many generations as 3D7 (see supplementary Figure 1). We are unaware of any commercially available anti-ATS antibodies, despite trying to find them. Several past publications have described such antibodies, but to date we have been unable to acquire them. We will continue to explore alternative sources for such antibodies, however as mentioned above, we believe our focus on coordination of var gene switching and a description of the central role var2csa plays in the var gene transcriptional network is a significant contribution to our understanding of antigenic variation in malaria parasites. The cell biological assays (growth rate comparisons and cell cycle progression, supplementary Figure 2 and Figure 5G) have been performed as requested.

We appreciate the reviewer’s comments on these details. We have greatly expanded the methods section and the figure legends to cover all the areas mentioned by the reviewer. We have also changed much of the annotation of the figures to make the experimental design and details clearer.

At the reviewer’s request, we have added an analysis of these data to Figure 1E and found that the differences are consistent and highly statistically significant. It is also worth noting that a similar analysis was derived from RNA-seq (Figure 4F), which gave a similar conclusion as the qRT-PCR data shown in Figure 1C and E. Thus, we are confident in the conclusions.

We have added cumulative data from a large number (74) of recently derived subcloned populations to demonstrate that the level of var transcripts observed in distinctly and consistently different. These data are now shown in the greatly expanded clone tree in Figure 1D and in the analysis of total var expression in Figure 1E.

As mentioned above, we do not currently have the ability to accurately assess protein expression levels but will attempt to do so in the future.

As mentioned above, the methodology we employed is limited to 3D7 and NF54, however we now cite other studies, including two that observed a similar phenomenon in the IT line (references 22 and 30).

Our choice of the 3D7 line is now explained in detail in the first section of the Results (see lines 127-134).

We have added a greatly expanded Methods section to describe in detail how we synchronized our cultures (see lines 589-598). This includes “double synchronization” to obtain very tight synchronization. Importantly, parasites are allowed to recover from the synchronization procedure for 18 or 38 hours prior to RNA extraction, thus enabling the recovery of consistent amounts of RNA.

We have added details to the methods section to cover the questions raised by the reviewer. All qRT-PCR reactions were indeed performed in a 384-well format, enabling duplicates or triplicates to be performed together in a single run. We now explain this in detail in the figure legends and methods section (613-615). We have added details regarding controls for the presence of gDNA and how much template as used for each individual reaction (see lines 601-604). The PCR primer set employed here was originally created specifically to ensure similar amplification efficiencies for each primer pair and to enable absolute quantification (see Salanti et al., 2003). It has been used extensively by numerous groups when investigating var gene expression in the 3D7/NF54 genetic background. We have added this to the methods section (see lines 604-608). For Figure 1B, the error bars represent standard deviation from three biological replicates (see legend to Figure 1B).

We have added the requested details to the Methods section (see lines 619-648), including an additional Supplementary table (Table S9). The analysis of gene families with multiple copies is not relevant since we omitted the multicopy gene families for the RNAseq analysis described in Figures 4 and 5.

We have corrected the text as requested.

As noted in our comments to reviewer 1, we have entirely renamed all the clonal lines to make it easier to determine their relationships to each other. We have also added a significantly larger number of clones to Figure 1D, including the 5 subclones shown in Figure 2A, to show exactly how the different parasite subclonal lines are related.

We have added the specific number of days in culture below each column of the heatmaps in Figures 2A and 4B to make the time of collection clear for each sample.

As also described in our response to reviewer 1, the heatmaps are no longer organized by hierarchical clustering. The genes are now arranged according to ups type.

An extensive analysis of var gene expression levels is now provided in Figure 1E. Indeed, throughout all the clones analyzed (74 clones), the expression level is highly correlated with the many vs single phenotype (p<0.0001).

These two clones have now been renamed “clone 4 and clone 5” to indicate that they are simply additional clones of 3D7 that were generated in our lab several years ago and cultured separately. They are thus derived from the same genetic background, but simply separated by time. This is now described in greater detail in the figure legend.

To avoid confusion, in the figure legend and the text we now refer to “subclone generations” and “replication cycles”. We hope this helps to avoid confusion.

We have added a much more extensive subcloning tree in Figure 1D, including subcloning of both “singles” and “manys”. It is clear from this experiment that either phenotype can be generated from a population that displays the alternative phenotype. It is true that in Figure 2B, each round of subcloning originated from a “single” line, however Figure 2C includes heatmaps from both “singles” and “manys”, with no apparent difference.

The cloning process takes approximately 5 weeks from isolation of an individual parasite until a population has expanded sufficiently for the analysis to take place. We have added this to the methods section (see lines 574-577). The heatmaps are now organized according to var type. The rows in the heatmap in Figure 2C are now organized from left to right to match the pie charts in Figure 2B, thus enabling a reader to determine this relationship. We have added a description of this to the figure legend.

In response to a previous suggestion from this reviewer, we have deleted the single cell analysis because of the limitations of this dataset.

This is an interesting idea. We have no reason to think this might be the case, particularly since var2csa-mutant clones can be either singles or manys, and thus they can progress either rapidly or slowly through the ring stage. Without a better hypothesis, we have chosen not to discuss this idea.

We have now organized all the heatmaps according to ups type and we don’t detect any correlation. However, these data are now available for readers to consider.

We thank the reviewer for pointing out that we did not clearly describe the different methods for disrupting or deleting var2csa or how parasites derived from each method were analyzed. This is now shown in the revised Figure 4 as well as described in the revised text (see lines 260-271).

We have added numerous references to this section to ensure the reader can easily access the relevant literature.

We have added the appropriate references to this sentence (see line 243).

We cite the Duffy papers in our discussion of var2csa promoter competition (see lines 458-461), where they fit nicely. We thank the reviewer for this suggestion.

We have incorporated these ideas and references into the Discussion section (see lines 475-478).

This is an interesting point of discussion that we were hoping to stimulate with the paragraph in question. We note that indeed we were able to detect very low levels of transcript by qPCR in our study, however we were using specific primers optimized for the parasite line used in our experiments. In contrast, in the case of the splenectomized patient, the investigators had to use degenerate primers since the genome sequence was not known, which might have reduced sensitivity below the threshold of detection. In addition, the patient is presumed to have extensive anti-PfEMP1 immunity, thus providing a heavy selection pressure against var gene expression, a selection that is missing in our cultures. We now mention the similar example from P. knowlesi and have added the Barnwell reference (see lines 533-536 and the accompanying text).

The text of the Figure has been enlarged to make it easier to read. We prefer to keep var1 as group D because this gene is unique amongst group A genes in that it does not recombine with other members of the family and represents an unusual gene that is conserved across species (see Gross et al., 2021). We have added the group designation next to the annotated pie charts as requested.

We have replotted all of the pie charts to adjust the colors so that the red color that represents var2csa is conserved throughout the manuscript. Only var2csa is represented by red.

To overcome the confusion from our original naming of the various clonal lines throughout the manuscript, we have renamed all clones in a systematic fashion. This also applies to the clones shown in Figure 4. This figure has also been reorganized to some extent for greater clarity.

As requested, we have added labeling of each var group and we indicate the dominantly expressed gene for each pie chart. We have also added the time of cultivation to each heatmap.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

To address this suggestion, as well as comment 4D below, we have assembled a complete clone tree that includes all 74 lines derived from the subcloning experiments. This is quite large and perhaps a bit cumbersome to view, therefore we have included it as Figure 1—figure supplement 1. To make this large tree easier to understand, we modified the numbering scheme slightly. Two clones that were not derived from the subcloning experiments but were used for comparison in Figure 2A are also displayed here. We prefer to keep the smaller clone trees in the main manuscript (Figures 1D and 2B) where they are easier to follow and illustrate specific points. We included stacked graphs of total var expression for each clone in the clone tree in Figure 1—figure supplement 1 as requested.

The question of how to systematically categorize a subcloned population as either “single” or “many” is both challenging and somewhat arbitrary, as the reviewer correctly points out. We believe that no population is entirely “single” or “many”, but rather is a mix dominated by parasites in one or the other state. To test whether “single” vs “many” populations have different total var transcript levels as we do in Figure 1E, it is important that we not include transcript levels in the definition. Therefore, we now categorize populations in which more than 50% of the total var transcript is derived from 1 or 2 var genes as being “single” while populations with more diverse expression patterns are considered “many”. This stricter, unbiased definition resulted in only changing the classification of one population in the complete clone tree shown in Figure 1—figure supplement 1 (out of 74). We also changed the classification of the population specifically mentioned by the reviewer, NF54-5 in Figure 1—figure supplement 2. This population barely meets the definition of a “single” and has a correspondingly low transcript level. We now explicitly explain our methodology and our interpretation of the results on lines 172-182 and 192-198 of the main text.

Indeed the “singles” catch up with the “manys” and replicate at the same overall rate. This is reflected in three sets of data: (1) the RNAseq data (Figure 5A) show that at the trophozoite stage, there is very little difference in gene expression (compared to the substantial changes observed in the ring stages), indicating that the two populations are at nearly identical points in the replicative cycle, (2) analysis by flow cytometry (Figure 5G) shows that there is no detectable difference in DNA content between “singles” and “manys” (a reflection of nuclear replication), thus overall progression of the replicative cycle in not discernably different, and (3) the overall replication rate is not different when comparing the different populations (Figure 5—figure supplementary 1). We now make this point more clearly on lines 472-477. We could add Giemsa-stained smears, however we feel that the data we’ve already included are more quantitative and less subjective than smears and thus smears would not add to the manuscript.

As mentioned previously, we are not mathematical modelers and would prefer to leave detailed discussions or the derivation of a new mathematical model to researchers who specialize in this methodology. We focused this manuscript on var2csa for the reasons detailed in the text, however we agree with the reviewer that it is possible that not all var genes are created equal. We have added to the discussion the possibility that other genes (and we specifically mention Pf3D7_0421100) could similarly serve as nodes in a hypothetical network (see lines 526-531). We thank the reviewer for this suggestion.

The labeling of the heatmaps have been modified to remove the redundancy of the var types, as suggested by the reviewer. We have also changed the order of the genes in the figures so that they all match.

We thank the reviewer for examining the figures so closely. There was indeed an error in assembling Figure 2A. We have corrected the time points and it is now clear that the times correlate between Figures 2A and 3A, although more time points are included in Figure 2. A later timepoint was included for clone 6 in Figure 3A because this more clearly shows convergence to var2csa expression. We now note that the time to convergence to var2csa is variable (see line 277), but all populations eventually express var2csa.

Once again, we thank the reviewer for catching this error. We now recognize that the population originally included in this figure (similar to the NF54 population discussed in response to point 1 above) was a poor choice given its initial expression profile, leading to confusion and mislabeling. To avoid confusion, we have replaced this population with an equivalent subclone in Figures 4A, 4E and 4F. This population was originally labeled V2dis3 and was included in the analysis in Figures 4E and 4F. We have now relabeled this as V2dis2 and show the pie chart for its var expression profile in Figure 4A. The population originally shown in Figure 4A is now labeled V2dis3 and is included in the analysis in Figures 4E and F. As expected, this population is the least polarized of the var2csa-mutant populations analyzed, although its var expression profile, as determined from RNA-seq, indicates it qualifies as a “single”.

The heat maps are designed to display the minor transcripts or “background” var expression pattern that we hypothesize might reflect switching biases. The heatmaps are presented so that readers can observe shifting patterns in the minor transcripts over time (Figure 2A) and when comparing closely related subclones (Figure 2C). var2csa is unique since it becomes activated in essentially all subcloned populations, thus we don’t consider it a “minor” transcript and it does not contribute to the purpose of these heatmaps. We now explain this in the text (see lines 228-230). We do indeed present expression data on var2csa, however we present it separately (see Figure 3A).

As requested, we now present a complete tree with all related subclones in Figure 1—figure supplement 1.

As requested, we now explicitly label var1 and the var3 variants, and include them as a subset of the group A var genes in Figures 1, 2, and 4.

The point of Figures 4C and D is to examine the differences between wildtype and var2csa-mutant lines, specifically by comparing how expression patterns shift over time. The data displayed in Figure 2C, while similar, does not reflect how the var pattern shifts within a population overtime, but rather how it shifts as populations go through the sequential bottlenecks of repeated subcloning. The analysis in Figures 4C and D thus reflects directly comparable experiments, while the data shown in Figure 2C is from a different experimental design. We now explain the nature of this comparison on lines 356-357. We are concerned that incorporating the additional, somewhat different experiment could be misleading. Further, adding another 5 sets of data points (44 additional dots) could obscure the message that we are trying to relate. We have therefore decided to leave the figure unchanged.

We checked for var2csa expression after disruption of the promoter using both Q-RT-PCR and RNA-seq and failed to detect any var2csa transcripts. We have added this confirmation to the text of the manuscript (see lines 303-305).