File S3

This .csv file contains all of the sequences used in the phylogenetic analysis (see above). For form annotation sequences under 360aa were excluded and no upper limit was set. Sequences removed by trimAL (using a gap threshold of 0.1) are labeled as “Unannotated” - many of them may not be actual rubiscos. Sequences that are too short are labeled as such. Some sequences will have a form indicated but not a subform, for instance, some sequences are labeled as Form III but with no indicated subform because they do not fit into an established subclade. We used a tree made from a 65% identity dereplication (using CD-HIT with standard parameters, File S4). The tree was produced as described above using IQTree with the following parameters: -bb 1000 -m MFP -safe. The tree was rooted just past the Form IIIA clade so that all bona fide rubiscos form one clade and all RLPs form another clade. There are a few branches in between that we consider to be RLPs.

There is a beta-hairpin sequence in most bonafide rubiscos that is absent from the Form IVs (DEAQGPFYR in R. rubrum). Erb and Zarzycki 2018 [5] implicate this structural feature in their argument regarding whether RLPs or Form III rubiscos came first. We find that this sequence is absent in Form IIIA sequences and is present in some but not all Form IIIC. This sequence is quite divergent between clades (e.g. Form II/III has an extended hairpin) and structurally they seem to vary a fair amount. There may be useful information in the phylogenetic distribution of this hairpin that may inform the placement of the root of the rubisco tree.

With one exception, clades were assigned by grouping together all branches that share a common ancestor with the reference sequences (see tab 2 of file S2). Then we assigned the same form to all sequences that were clustered together with the CD-HIT algorithm.

The A. fulgidus clade could not be assigned in this way. This paraphyly is apparent in [6]. In order to overcome this obstacle we remade the tree at 70% identity dereplication (File S5). We also used 5 sequences from figures S12 and S13 of Erb et al. 2012 in order to pinpoint the A. fulgidus clade. Four of the Five clustered into a single clade while the fifth (ZP_08130208.1, sister to WP_025656390.1) branches much closer to the YkrW clade. In order to avoid mislabeling we have omitted the clade containing that reference and assign the A. fulgidus clade with just the other four references (WP_010879084.1, ZP_09117828.1, MBQ5951511.1, and WP_012813926.1).

Form IE rubiscos may be paraphyletic, sequences were chosen based on clades containing reference sequences from File S2 tab2.

Sequences that diverge before or after the Form IEs are labeled as Form I with no subform specified. Similarly, there are a few outgroups to the Form IA and IB that are not assigned a subform.

Form ID emerges from within the Form IC: the distinction is taxonomic and not phylogenetic, Form ID rubiscos are eukaryotic while IC are prokaryotic.

We include some additional sequences in the III-like clade because they branch very closely to the remainder of the III-like and far from everything else.

Form I is defined as all sequences in a clade containing the Form I alphas. This excludes one sequence that is between the Form I alphas and the Form III transaldolase variants: MCA9846407.1

Form I'' can diverge before or after Form I' depending on the tree model used. In Schulz et al. 2022 [32] they constrain it to diverge after I' but with a bootstrap of just 23 (they used model LG by using a best fit approach in RaxML). This constraint was imposed because of a short insertion common to all Form I sequences and Form I’’ that is absent in Form I’ and Iα - parsimony would clearly indicate that Form I’’ is more derived (see Schulz et al. 2022 supplementary text).

When we make the tree using the model from West-Roberts et al. 2021 [16] (LG+F+G, File S6) we get a similar result to Schulz et al., with I'' diverging after I' and a bootstrap of 29. When we used model finder in IQTree it chooses LG+R8 and we get a bootstrap of 99 with I' diverging before I'' (File S7). The placement of the I'' clade is therefore sensitive to the exact sequences used and the alignment model. With advances in metagenomics and the discovery of new sequences of this enigmatic group, the resolution of the position in the tree may improve.