Predicted loss and gain of function mutations in ACO1 are associated with erythropoiesis.

Hemoglobin is the essential oxygen-carrying molecule in humans and is regulated by cellular iron and oxygen sensing mechanisms. To search for novel variants associated with hemoglobin concentration, we performed genome-wide association studies of hemoglobin concentration using a combined set of 684,122 individuals from Iceland and the UK. Notably, we found seven novel variants, six rare coding and one common, at the ACO1 locus associating with either decreased or increased hemoglobin concentration. Of these variants, the missense Cys506Ser and the stop-gained Lys334Ter mutations are specific to eight and ten generation pedigrees, respectively, and have the two largest effects in the study (EffectCys506Ser = -1.61 SD, CI95 = [-1.98, -1.35]; EffectLys334Ter = 0.63 SD, CI95 = [0.36, 0.91]). We also find Cys506Ser to associate with increased risk of persistent anemia (OR = 17.1, P = 2 × 10-14). The strong bidirectional effects seen in this study implicate ACO1, a known iron sensing molecule, as a major homeostatic regulator of hemoglobin concentration.


Point-by-point answer to reviewers
We would like to thank the three reviewers for their critique. We have made substantial changes to the manuscript in accordance to their comments and believe that we have addressed all their concerns.
All changes to the originally submitted manuscript can be reviewed in track-changes. Line numbers below refer to edits in the manuscript with track changes on.

Response to Comment #1 a
In the current study we chose to focus on the biological effects of several variants in a single gene. It is unusual to observe several naturally occurring protein altering variants in the same gene with such a large opposing effects. These unique results at the ACO1 locus deserve to be presented fully, and would not be done justice as a subsection in a large GWAS study cataloging all hemoglobin associating variants.
To address the reviewers concerns we will share summary statistics upon publication for GWAS of hemoglobin concentration in the Icelandic population. Nor an explanation of why it was then focused on at a subthreshold level (e.g., rs147876514 has p=9e-04), nor any discussion on the implications of doing things this way.

Response to Comment #1 b
There were five variants at ACO1 associating with hemoglobin concentration at genome-wide significance, of which three are coding variants (Cys506Ser, Arg168Trp, and Thr208Ala) and two were common non-coding (rs12985, rs7045087). It is quite unusual to identify multiple independent variants in the same gene, especially when these variants are coding and have large opposing effects on a trait. Collectively these variants strongly implicate ACO1 as an important regulator of hemoglobin concentration. Thus, we performed a secondary analysis to test all coding variants in the ACO1 gene, and we chose a P-value threshold for the secondary analysis based on a Bonferroni correction for the number of coding variants in ACO1 (N=34).
To address the concerns of the reviewer, we have added the following text to the introduction and discussion (changes are underlined).  (Table 1). Subsequently, we tested the 34 remaining coding variants in ACO1 for association with hemoglobin concentration and found three additional associations after accounting for multiple testing (P < 0.05/34 =

x 10 -3 ) (Table 1 and Supplementary Data 1). "
Page 9, lines 228-230 in the discussion: "The aim of this study is to understand how sequence variants in ACO1 affect hematopoiesis. After we identified a genome wide significant association, we performed conditional analysis to identify secondary associations at the locus, focusing on variants with a predicted protein coding effect on ACO1. " 4

Comment #2
Did you test previously reported hits for replication?

Response to Comment #2
We have added information on all reported GWAS associations with hemoglobin levels and how they replicate in Iceland and the UK Biobank. This information is now provided as supplementary material (supplementary data 7).

Response to Comment #3
To estimate the heritability of hemoglobin concentration, we calculated the SNP-heritability using LD score regression. We have now included these results as a supplementary table (Supplementary Table 2) and corresponding methods (Changes underlined).  Table 2)." And page 14, a new subsection at lines 372-375

"Heritability
Heritability of hemoglobin concentration was estimated in the following two ways: 1) 2 × parent-offspring correlation, 2) 2 × full sibling correlation, using the Icelandic data (where all family relationships are known)."

Comment #4
Methods state that the LD-Score correction factor was 0.68 in Iceland and 1.40 in UK Biobank. Is this the inflation factor (if not, also report the raw inflation factors), or the actual amount it was estimated that it needed to be corrected for? Much more description of what is going on here might be helpful. Why in particular is the Iceland one so off? The UK Biobank might be reasonable for a very polygenic trait. Are you trying to say that you deflated the UKB statistics, but you actually inflated the Iceland statistics?

Response to Comment #4
When testing association of sequence variants with quantitative traits we use a BOLT linear mixed model. These models are now widely used because they account for cryptic relatedness while also increasing power 6 .
One-step in the BOLT-LMM procedure (step 1b) is to calibrate the 2 test statistic by calculating a constant calibration factor cinf in To compute the calibration constant BOLT-LMM rapidly computes the prospective statistic at 30 random SNPs by applying conjugate gradient iteration.
We, however, did not apply this scaling to the test statistic in the Icelandic association model. Therefore, when we apply the LD score regression and estimate a correction factor from the regressions intercept it will be shifted by this constant factor. The intercept is therefore not comparable to factors obtained from standard genomic control methods, and should not be interpreted as such. It can indeed be below 1 due to the calibration factor.
If associations were computed under a standard generalized linear regression model the correction factor for the Icelandic data would be 1.57 which is like the reviewer points out reasonable for a very polygenic trait. We are therefore not inflating the Icelandic statistics when 7 applying genomic control. The text from the methods section where we state that the correction factor is 0.68 has now been removed in order to avoid confusion.

Comment #7
Methods very cursory. Explain how did you trace the variants back to common ancestor born in the 18th century?

Response to Comment #7
Close to complete genealogical records of the Icelandic population are available dating back to the Icelandic national census of 1703, and incomplete records dating back to the settlement of Iceland in 874 CE 9,10 . The Icelandic genealogy coupled with the large fraction of the population that has been chip typed allows the determination of the origin of sequence variants of chiptyped individuals through long-range phasing and haplotype imputation 11 .
All carriers of the two rarest ACO1 sequence variants, Cys506Ser and Lys334Ter, can be traced back to common ancestors in the genealogical records by identifying the most recent common ancestor of all carriers of each variant in the genealogical record.
Also, these rare variants are not found on the same haplotype background in any descendants of relatives of the common ancestor. Therefore it is highly likely that each of the mutations originated from a single common ancestor.
A description has now been added to Methods section Page 13, lines 340-348.

Comment #10
What, if anything, did the alternative prioritization gain you?

Response to Comment #10
The alternative prioritization gained us three protein coding variants in ACO1, in addition to the two genome-wide significant ones.
These five protein coding variants allowed us to put ACO1 protein alterations into the context of existing literature on molecular biology of ACO1 based on in cell-and animal models.
Specifically, in the case of the Cys506Ser missense mutation we had the opportunity to study in humans the effects of a mutation that had been experimentally introduced into model systems to study the function of ACO1.

Comment #1
The paper would be improved by a better description of the switch from cytosolic aconitase to IRE binding. This is well described in Rouault, 2006, Nature Chem Biol.

Response to Comment #1
We thank the reviewer for this comment and we have now added the following text to the discussion chapter (changes underlined)

Response to Comment #2
We thank the reviewer for pointing this out. The wording on page 5 has been made clearer in accordance to the reviewers suggestions (changes underlined).

Response to Comment #3
We thank the reviewer for pointing this out. Walden and Volz (2006) is now referenced as suggested and is now reference no. 25 and is cited at page 7, line 181.

Response to Comment 4#
We thank the reviewer for pointing this out.

Response to Comment 5#
The point that the reviewer is making is interesting and we now discuss it in the manuscript.
We have added the following text in the discussion (changes underlined).

Comment #6
This is a very interesting paper that would benefit from a better narrative about why one protein could have two opposing roles, causing either anemia or polycythemia, depending on the mutation. it is also interesting that these mutations adversely affected heterozygotes.

Response to Comment #6
We thank the reviewer for pointing this out. To make these messages clearer for the reader we have changed the final paragraph in the Discussion section, It was not easy to recognize which of the patients and mutations mentioned were found in homozygotes. The information was in the paper, but not readily accessible.

Response to Comment #7a
To address the concerns of the reviewer we have made a

Response to Comment #7b
To address the concerns of the reviewer we have remade Figure 1 and Figure    All 67 genotyped carriers can be traced back to a common ancestor in the early 18th century. Shown is a shortest path pedigree. The founding couple had six offspring two of which were carriers of the Lys334Ter variant, and a current total number of 21,423 descendants. Roman numerals indicate generation, year of birth of the founding couple is noted above the symbols and mean hemoglobin concentration is noted below the symbols. square = male, circle = female, diamond = sex unspecified, solid filled object = carrier, half filled object = obligate carrier, red filled object = polycythemic. Obligate carrier status is not indicated before generation VII as no phenotype information is available for those individuals.

Comment #8a
The authors did a good job tying their findings to the discoveries in the literature, but some of the tables are quite oriented towards geneticists and would benefit from longer and clearer legends. Supplementary figures 4 and 5 could be more clearly presented.

Response to Comment #8a
We thank you for pointing this to us. We have now make changes to table legends for the following tables and they now read (changes underlined):   28

Comment #8b
Supplementary figures 4 and 5 could be more clearly presented.

Response to Comment #8b
For Supplementary figure 4. the figure the font size has been increased and the unit of measurement has been added to the axis labels.

Comment #1a
The main focus is on rare variants within one locus but it is not clear why they are reporting only variants at or nearby this gene.

Response to Comment 1#a
We chose to focus on the biological effects of several variants in ACO1. It is unusual to observe several naturally occurring protein altering variants in the same gene with large opposing effects. These unique results deserve to be presented fully, and would not be done justice as a subsection in a large GWAS study cataloging all hemoglobin associating variants.
See response to Reviewer #1 comment #1a for a more comprehensive response 31

Comment #1b
This also raises questions about the overlap of the reported results with previously published studies or other ongoing studies using these datasets.

Response to Comment #1b
The Icelandic dataset has not been used in previous GWAS of hemoglobin concentration.
A fraction of the UK biobank dataset was used in the study by Astle  The only previously reported marker reported in the current study is the common variant rs12985 1 .
It is however difficult to speculate on other studies. Clearly the UK Biobank is public and researchers can freely access that data.
To demonstrate that the datasets used in the current study are consistent with previous reports we now provide replication of previous reports (see response to Reviewer #1 comment #2 ).
In addition, to allow other researchers to benefit from our efforts we will share summary statistics down for hemoglobin concentration upon publication.

Comment #2
There is also some lack of clarity on the choice of the significance thresholds for p-values (including multiple thresholds) and how the study went from a discovery using combined Iceland/UK Biobank datasets to single study results for the ACO1 gene (Table 1).

Response to Comment #2
See response to Reviewer #1 comment #1b 33

Comment #3
Line 78, which common variant in Table 1 is the one that captures previously reported intergenic association and which methods were used to get to this conclusion.

Response to Comment #3
The variants rs7045087 is the same variant that was reported by Astle at al. at this locus. We came to this conclusion by referencing the GWAS catalog database and checking the original publication.
• The text in lines 80-81 (previously line 78) now reads "...of which three are coding and one common non-coding variant rs7045087 represents a previously reported intergenic association (Table 1)." • In table 1 the reported variant rs7045087 is marked with a footnote (十) that reads "Previously reported in Astle, 2016".

Comment #4
Line 91, variants were also associated with RBC and HCT, which is not surprising as these are highly correlated phenotypes.

Response to Comment #4
We agree with the reviewer that hemoglobin concentration, RBC and HCT are all highly

Response to Comment #5
In order to investigate iron storage, we tested the variants for association with iron, ferritin, iron binding capacity, and transferrin saturation in the Icelandic population (

Comment #6
Paragraph starting in line 114, the title is misleading given the lack of functional studies for variants and assumptions are based on the direction of the beta estimates.

Response to Comment #6
In case the of the Cys506Ser variant, the mechanism is well studied in model systems and is the result of a gain-of IRE-binding. For the other coding variant, Thr208Ala, which shows the same direction of effect, we are making the assumption that it is going through the same mechanism of action based on the known mechanism of the Cys506Ser variant. In the text we are careful to state that this mechanism of action is suggestive and is based on effect estimates.
We have changed the title to (changes underlined), page 5, line 119: "Predicted RNA binding gain-of-function variants"

Comment #7
For the sex-specific results, report the number of individuals with the alternative allele for rare variant within sex.

Response to Comment #7
We thank the reviewer for pointing it out. We have adjusted the text by adding the suggested information.

Response to Comment #8
The study is based on a large set of individuals who represent a large fraction of a founder population. Our method of testing for association takes the closest relatedness into account using a mixed effect model implemented in BOLT-LMM 6 .
See also Response to Reviewer #2 Comment #4 and the "Association analysis" subsection in the

Response to Comment #9
We performed a phenome-wide scan testing the Cys506Ser missense variant in ACO1 for association with 396 binary phenotypes in Iceland. No significant association with binary phenotypes was observed in addition to persistent anemia as reported in the manuscript when taking the number of tested phenotypes into account ( P < 0.05/396 = 1.3 x 10 -4 ) .
We have now adjusted the text by adding the suggested information.

"Consistent with the large effect on hemoglobin concentration, we detect an association of
Cys506Ser with a high risk of persistent anemia (all hemoglobin measurements < 118 g/L for women and < 134 g/L for men) (

Comment #10
Any evidence for selection at the region?

Response to Comment 10#
To our knowledge the ACO1 locus has not been reported to be under selection 16 .
In the gnomAD database ACO1 does not show signs of selective constraint as measured by pergene constraint scores. ACO1 has a Z-score of 0.92 for missense variants that indicates that the observed count of missense variation in the gene does not deviate significantly from what would be expected under neutral selection. The pLI (probability of loss-of-function intolerance) score for ACO1 is 0 indicating that loss-of-function variation in the gene is tolerated. This data is consistent with our observations that coding variation in ACO1 has moderate impact on health and is unlikely to affect reproductive success. At least at the heterozygous level the rarest variants with the largest effects, Cys506Ser and Lys334Ter, are observed within extended pedigrees, and are thus compatible with reproduction. However, we observed no homozygotes for these variants which is consistent with their rarity and we can therefore not conclude anything when it comes to the effect of these variants on homozygotes.

Comment #11
Line 265, give a more concrete example on how drug therapy targeting ACO1 can help in anemia due to chronic inflammation (see above question).

Response to Comment #11
As we have substantially deepened the discussion on the biological and genetic effects of ACO1 variants (see reponses to Reviewer #2, comments #1 & #6) we have decided to delete the sentence on anemia of inflammation ("Many chronically anemic patients, such as those with anemia of inflammation, do not respond well to EPO"). We furthermore concede that this suggestion is highly speculative and should thus be deleted.

Comment #12
Line 285, for the multiple whole genome sequencing deCODE projects, were the datasets called and qc together?

Response to Comment #12
The whole-genome sequenced samples were variant called jointly and the sequence variants found through whole-genome sequencing were phased jointly. The long-range phasing and imputation steps are performed for all chip-typed individuals, participating in various disease projects at deCODE Genetic, simultaneously 11 . This approach incorporates many different quality controls to overcome batch effects and provides accurate genotype-calling -in

Comment #13
Line 289, more details on the imputation are needed (minimal allele frequency, imputation quality).

Response to Comment #13
Details on the imputation have been added to the sentence.
Page 12, lines 326-328 now read (changes underlined): "The chip-typed individuals were long range phased, and the variants identified in the whole-genome sequencing of Icelanders were imputed into the chip-typed individuals (imputation info > 0.8 and MAF > 0.01%)."

Comment #14
Line 300, sample size for the UK Biobank does not match the sample used in this analysis. We also note that effect on hemoglobin concentration reported in the current study are adjusted for age.

Comment #15b
Can you also show the variance of the hemoglobin over time across these groups? Same for the iron measures including ferritin.

Response to Comment #15b
For clarity we have also added Supplementary

Response to Comment #16a
In Iceland the hemoglobin concentration measurements were obtained from hospital records independent of patient status. For patients with multiple measurements we used the mean value in the analysis. The nature of the data and its origin is such that most measurements are from ambulatory setting, since hemoglobin is always measured in blood tests. We adjusted for time to death to correct for anemia because of illness. We used the mean value for those with multiple measurements in the analysis, which should decrease the effects of illnesses and hospitalizations on hemoglobin concentration.

Comment #16b
Why adjust for country of origin?

Response to Comment #16b
Iceland is historically divided into 23 counties corresponding to different geographical regions of the island (https://en.wikipedia.org/wiki/Counties_of_Iceland). We adjust for county (i.e. not country) of origin within Iceland which would take geographical variation in sequence diversity in Iceland into account.
See response to comment 16c# bellow for a more detailed discussion on the topic.

Comment #16c
Principal components were not included in analyses and adjustments for relatedness are not described.

Response to Comment #16c
In the Icelandic population the first principal components correlate strongly with the county of birth 17,18 . However, they account for a very small proportion of genotypic variance 18 . This suggests that the effect of adjusting for these components will be very small. For the Icelandic dataset county of origin is included as a covariate in the logistic regression model and corrected for the QT analysis 6 (See "Association analysis" subsection in Methods).
For information on adjustments for relatedness see response to Reviewer #3 comment #8 For information on the UK dataset see the "Association analysis" subsection in the Methods page 14, lines 386-387:: "In the UK Biobank study, 40 principal components were used to adjust for population stratification and age and sex were included as covariates in the logistic regression model and the BOLT-LMM." 49 Comment #17

Supplementary Table suggest differences in the distribution of hemoglobin between
Iceland and UK biobank. Given Iceland used multiple measures and UK biobank just one measure, are the scale of estimates equivalent for a meta-analysis using inverse variance methods?

Response to Comment #17
The Icelandic measurements were standardised prior to taking the average over multiple measurements, the same way as for the UK biobank data. Hence the measurements in Iceland and UK biobank are on the same scale (in normalized standard deviations). However, as shown in Table 1, the variance of the measurements is larger in the Icelandic data compared the UK biobank data. The reason is probably that most of the Icelandic measurements come from hospitals and therefore include many patients that have abnormal hemoglobin levels. As the distributions are standardized, this would affect the effect estimates, leading to lower effect estimates in the Icelandic GWAS.

Comment #18
Significance threshold paragraph (line 354) is not clear.

Response to Comment #18
Significance threshold paragraph now reads  N = 35,567,755)." As you have written in the response, can you explicitly say in the manuscript "It is quite unusual to identify multiple independent variants in the same gene, especially when these variants are coding and have large opposing effects on a trait.", and then say that is why you focused on the ACO1 gene?
Need CIs or some assessment of error on heritability estimates in text.
Reviewer #2 (Remarks to the Author): In this resubmission, the authors have greatly improved the narrative and readability. The findings are very interesting and important. This reviewer could not assess the changes in the manuscript are they were not labelled. Based on the answer to reviewers' questions, the revised manuscript addressed some the issues raised but further clarifications are still needed. There is still lack of clarity on the focus of the paper. The introduction says "we focus on the rare missense and loss-of-function variants with large effects on hemoglobin concentration." but the paper just describes the ACO1 findings.
Answer to R#3, question 1b. for replication, please remove the overlap samples UK Biobank). R#3, question 2. Answer to R#1 comment 1b does not include the steps leading to focus on the ACO1 gene. Perhaps a figure with the steps for the study design can clarify this given both R#1 and R#3 have the same questions. If variants at the ACO1 gene were the only significant findings of the study, then state this in the result section. Adding Manhattan and QQ plots for discovery would be helpful to clarify this. The study should also report lambdas for discovery studies. R#3, question 3. Did you use conditional analysis for this, if so, describe in methods. R#3, question 16a. I don't think adjusting for age at death is the appropriate adjustments to do. I noticed that the justification for this variable is not included in the revised manuscript. There are concerns that some of the measures were obtained during illness which can bias the results (anemia due to blood dilution, acute illnesses, bleeding). The authors need to either remove measures from hospitalizations (just use ambulatory measures) and/or remove measures close to time of death. This is relevant considering answer to question #17 that states that most of the measures are from hospitalizations. Table 1. include the minor allele count for low frequency variants. Results, lines 84-85, sentence "Subsequently, we tested the 34 remaining coding variants in ACO1 for association with hemoglobin concentration and found three additional associations after accounting for multiple testing". The rationale for the strategy including the selection of the variants and choice of significance thresholds needs further support. This is not described in methods.
For Icelandic study, please state if informed consent was obtained.

Point-by-point answer to reviewers -second review
We would like to thank the three reviewers for their critique in the second review. We have made substantial changes to the manuscript in accordance to their comments and believe that we have addressed all their concerns.
All changes to the manuscript from previous review can be reviewed in track-changes. Line numbers below refer to edits in the manuscript without track changes..

Reviewer #1 (Remarks to the Author):
Comment #1 As you have written in the response, can you explicitly say in the manuscript "It is quite unusual to identify multiple independent variants in the same gene, especially when these variants are coding and have large opposing effects on a trait.", and then say that is why you focused on the ACO1 gene?

Response to Comment #1
To make this message clearer for the reader we have now emphasized that it is unusual to identify multiple independent variants in the same gene and give a rationale for the focus on

Comment #2
Need CIs or some assessment of error on heritability estimates in text.

Response to Comment #2
Confidence intervals on heritability estimates as an assessment of error have been added to the text. This reviewer could not assess the changes in the manuscript are they were not labelled.
Based on the answer to reviewers' questions, the revised manuscript addressed some the issues raised but further clarifications are still needed. There is still lack of clarity on the focus of the paper.

Comment #1
The introduction says "we focus on the rare missense and loss-of-function variants with large effects on hemoglobin concentration." but the paper just describes the ACO1 findings.

Response to Comment #1
Regarding other findings of the hemoglobin GWAS meta-analysis in Iceland and UK, in the main text we now mention the total number of loci with genome-wide significant signals, the replication of known hits, and the number of loci with genome wide significant rare coding variants. Subsequently, we underline the uniqueness of the finding at the ACO1 locus. We also provide supplementary data summarising these findings

Response to Comment #3a
We thank the reviewer for pointing this out. To make these messages clearer for the reader we now include a flowchart of the study design that describes the steps leading to the focus on the ACO1 locus (Supplementary Figure 1).
See also the response to comment #1 from Reviewer#1 Supplementary Figure 1

Response to Comment #3b
A statement has now been added to the Result section regarding additional findings of the current hemoglobin GWAS meta-analysis. We also provide a manhattan plot for the metaanalysis (Supplementary Figure 3), and QQ-plots of the GWAS results in Iceland and the UK (Supplementary Figure 11). Also, summary statistics are provided for the GWAS metaanalysis of hemoglobin levels for all tested variants in Iceland and the UK datasets.
See also the response to comment #1 from the same reviewer R#3, question 16a. I don't think adjusting for age at death is the appropriate adjustments to do. I noticed that the justification for this variable is not included in the revised manuscript.

Response to Comment #5a
Age at death is not used as a covariate in the linear mixed model when testing for the association of sequence variants with quantitative traits. For binary phenotypes, current age is used as a covariate for living individuals in the logistic regression model, and alternatively age at death is used for the deceased.
To make this clear to the reader we have made changes to the subsection "Association models" in the "Online methods" section of the manuscript

Comment #5b
There are concerns that some of the measures were obtained during illness which can bias the results (anemia due to blood dilution, acute illnesses, bleeding). The authors need to either remove measures from hospitalizations (just use ambulatory measures) and/or remove measures close to time of death. This is relevant considering answer to question #17 that states that most of the measures are from hospitalizations.

Response to Comment #5b
The studied populations in Iceland and UK are quite different, however when testing reported variants we observe similar effect sizes.
In Iceland, the median age of measurement is 72 years of age, whereas it is 60 in the UK.
Participants had on average over 6 measurements in Iceland (geometric mean = 6.4), we used the mean value for those with multiple measurements in the analysis. Also, we used the raw hemoglobin concentration measurements to derive anemia and polycythemia status when measurements were always above or below their respective diagnostic thresholds. In the UK, a single measurement was performed on blood samples obtained from UK Biobank assessment centre visit. Thus, the differences in the raw hemoglobin measurements between the two populations are a function of the differences in recruitment practices mentioned above (Supplementary Table 1 and Supplementary Figure 2).
The main aim of this paper is discovering associations of rare coding sequence variants with a large effect on hemoglobin concentration. To demonstrate robustness, in Icelandic and UK datasets we replicate the large majority (129 of 131) of hemoglobin associated variants reported in European populations. Also, we compared effects in standardized and raw scale (g/L) for the 131 hemoglobin associated variants reported in European populations to explore whether there is a difference in effect estimates between the Icelandic and UK datasets ( Supplementary Fig. 10). There are 27% higher effect estimates on the standardized scale in the UK dataset than in the Icelandic one (ratio of effect UK/Iceland = 1.27 (95% CI 1.23-1.32)). We note that the variance of raw hemoglobin concentration is higher in the Icelandic dataset than the UK one (standard deviation of raw hemoglobin concentration: Iceland = 15.5 g/L, UK = 12.2 g/L) (Supplementary Table 1). Once effect estimates are converted to raw scale (g/L) the effects are almost identical in the Icelandic and UK datasets (ratio of effect UK/Iceland = 1.02 (95% CI 0.99-1.06)) ( Supplementary Fig. 10). Thus, it appears that the difference in effect estimates on the standardized scale between UK and Iceland can largely be explained by the higher variance in hemoglobin concentration in Iceland.
For variants at the ACO1 locus, no significant difference between the effects in Iceland and UK on hemoglobin levels is observed (heterogeneity P-value < 0.05/4 = 0.0125) ( Table 1). In addition, when testing the variants Cys506Ser and Lys334Ter at ACO1, which are only present in Iceland, we presented the raw values of carriers and non-carriers stratified by age and sex and observe no indication that the effects are limited to sex or a specific age group (Supplementary Figure 7).
The UK and Iceland datasets included in the present analysis are diverse in regard to recruitment practices, but we still observe similar effects of sequence variants in the datasets. Despite the fact that the distribution of raw hemoglobin values is different in Iceland and the UK, the effect of the reported variants is of similar size in standardized values.
To emphasize the robust replication of previously reported hemoglobin associated variants and consistency of effect between the Icelandic and UK datasets in the current study we have made changes to the "Reported hemoglobin associated variants" subsection in the Results. Also, we have added a scatter plot comparing the effect of reported hemoglobin associated variants between the Iceland and UK datasets in standardized and raw values ( Supplementary Fig. 10).  Supplementary Fig. 10). For the combined Icelandic and UK datasets 129 out of 131 variants replicate. We also compared effects in standardized and raw scale (g/L) for the 131 hemoglobin associated variants reported in European populations to explore whether there is a difference in effect estimates between the Icelandic and UK datasets (Supplementary data 2 and Supplementary Fig. 10). There are 27% higher effect estimates on the standardized scale in the UK dataset than in the Icelandic one (ratio of effect UK/Iceland = 1.27 (95% CI 1.23-1.32)). We note that the variance of raw hemoglobin concentration is higher in the Icelandic dataset than in the UK one (standard deviation of raw hemoglobin concentration: Iceland = 15.5 g/L, UK = 12.2 g/L) (Supplementary Table 1). Once effect estimates are converted to raw scale (g/L) the effects are almost identical in the Icelandic and UK datasets (ratio of effect UK/Iceland = 1.02 (95% CI 0.99-1.06)) ( Supplementary Fig. 10). Thus, it appears that the difference in effect estimates on the standardized scale between UK and Iceland can largely be explained by the higher variance in hemoglobin concentration in Iceland.
The UK and Iceland datasets included in the present analysis are diverse in regard to recruitment practices 9,29 . Despite differences in age, population coverage, number, and purpose of measurements between the Icelandic and UK datasets, that are reflected in differences in the distribution of raw hemoglobin values (Supplementary Table 1 and Supplementary Fig. 2), we still observe similar effect of sequence variants on hemoglobin concentration in the two datasets (Table 1

Response to Comment #6
Minor allele count has been added to Table 1.

Comment #7
Results, lines 84-85, sentence "Subsequently, we tested the 34 remaining coding variants in ACO1 for association with hemoglobin concentration and found three additional associations after accounting for multiple testing". The rationale for the strategy including the selection of the variants and choice of significance thresholds needs further support. This is not described in methods.

Response to Comment #7
We have added a description of the gene-based strategy to the Online Methods section.